Models - Alibaba Cloud Model Studio - Alibaba Cloud Documentation Center

Flagship models

Global

In Global deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are dynamically scheduled globally.

Flagship models	Qwen-Max Ideal for complex tasks. The most powerful model.	Qwen-Plus A balance of performance, speed, and cost.	Qwen-Flash Ideal for simple jobs. Fast and low-cost.	Qwen-Coder An excellent code model that excels at tool calling and environment interaction.
Max context window ^(Tokens)	262,144	1,000,000	1,000,000	1,000,000
Min input cost ^{(Million tokens)}	$1.2	$0.4	$0.05	$0.3
Min output cost ^{(Million tokens)}	$6	$1.2	$0.4	$1.5

International

In International deployment mode, the access point and data storage are located in the Singapore region, and inference computing resources are dynamically scheduled globally (excluding Mainland China).

Flagship models	Qwen-Max Ideal for complex tasks. The most powerful model.	Qwen-Plus A balance of performance, speed, and cost.	Qwen-Flash Ideal for simple jobs. Fast and low-cost.	Qwen-Coder An excellent code model that excels at tool calling and environment interaction.
Max context window ^(Tokens)	262,144	1,000,000	1,000,000	1,000,000
Min input cost ^{(Million tokens)}	$1.2	$0.4	$0.05	$0.3
Min output cost ^{(Million tokens)}	$6	$1.2	$0.4	$1.5

US

In US deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are limited to the US.

Flagship models	Qwen-Plus A balance of performance, speed, and cost.	Qwen-Flash Ideal for simple jobs. Fast and low-cost.
Max context window ^(Tokens)	1,000,000	1,000,000
Min input cost ^{(Million tokens)}	$0.4	$0.05
Min output cost ^{(Million tokens)}	$1.2	$0.4

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Flagship models	Qwen-Max Ideal for complex tasks. The most powerful model.	Qwen-Plus A balance of performance, speed, and cost.	Qwen-Flash Ideal for simple jobs. Fast and low-cost.	Qwen-Coder An excellent code model that excels at tool calling and environment interaction.
Max context window ^(Tokens)	262,144	1,000,000	1,000,000	1,000,000
Min input cost ^{(Million tokens)}	$0.459	$0.115	$0.022	$0.144
Min output cost ^{(Million tokens)}	$1.836	$0.287	$0.216	$0.574

Model overview

Global

In Global deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are dynamically scheduled globally.

Category	Subcategory	Description
Text generation	General-purpose large language models	Qwen large language models: Commercial models (Qwen-Max, Qwen-Plus, Qwen-Flash), open source models (Qwen3)
	Multimodal models	Visual understanding model Qwen-VL
	Domain-specific models	Code model, Translation model
Image generation	Text-to-image	Wan text-to-image: Basic text-to-image: Generates beautiful images from a single sentence. Mixed text and image outputs: Generates a text description followed by a corresponding image, resulting in a seamless text-and-image output.
Image generation	Image editing	Wan image editing: Supports scenarios such as multi-image fusion, style transfer, object detection, image restoration, and watermark removal. Model series: Wan2.6.
Video generation	Text-to-video	Generates high-quality videos with rich styles from a single sentence.
	Image-to-video	First-frame-to-video: Uses an input image as the first frame and generates a video based on a prompt.
	Video-to-video	Reference-to-video: Generates a video that maintains character consistency using a prompt and the appearance and voice from an input video.

International

Category	Subcategory	Description
Text generation	General-purpose large language models	Qwen large language models: Commercial models (Qwen-Max, Qwen-Plus, Qwen-Flash), open source models (Qwen3, Qwen2.5)
	Multimodal models	Visual understanding model Qwen-VL, visual reasoning model QVQ, omni-modal model Qwen-Omni, and real-time multi-modal model Qwen-Omni-Realtime
	Domain-specific models	Code model, Translation model, Role-playing model
Image generation	Text-to-image	Qwen-Image: Excels in handling complex instructions, rendering both Chinese and English text, and generating high-definition, photorealistic images. It supports the selection of different models based on efficiency and quality requirements. Wan text-to-image: Basic text-to-image: Generates beautiful images from a single sentence. Mixed text and image outputs: Generates a text description followed by a corresponding image, resulting in a seamless text-and-image output. Tongyi - text-to-Image - Z-Image: A lightweight text-to-image model that quickly generates high-quality images and supports bilingual rendering in Chinese and English, complex semantic understanding, and a variety of styles and themes.
Image generation	Image editing	Qwen-Image-Edit: Supports prompts in both Chinese and English, enabling complex image and text editing operations such as style transfer, text modification, and object editing. It also supports multi-image fusion and is adaptable to a wide variety of industrial application scenarios. Wan image editing: Supports scenarios such as multi-image fusion, style transfer, object detection, image restoration, and watermark removal. Model series include the following: Wan2.6, Wan2.5.
Speech synthesis and recognition	Speech synthesis (text-to-speech)	Qwen speech synthesis and Qwen realtime speech synthesis can be used for text-to-speech in scenarios such as intelligent voice customer service, audiobooks, in-car navigation, and educational tutoring.
Speech synthesis and recognition	Speech recognition and translation	Qwen realtime speech recognition, Qwen audio file recognition, Qwen3-LiveTranslate-Flash-Realtime, and Fun-ASR speech recognition can perform speech-to-text for scenarios such as real-time meeting records, real-time live stream captions, and telephone customer service.
Video generation	Text-to-video	Generates high-quality videos with rich styles from a single sentence.
	Image-to-video	First-frame-to-video: Uses an input image as the first frame and generates a video based on a prompt. First-and-last-frame-to-video: Generates a smooth and dynamic video based on the provided first and last frames and a prompt. Multi-image-to-video: Generates a video by referencing the entity or background in one or more input images, combined with a prompt.
	Video-to-video	Reference-to-video: Generates a video that maintains character consistency using a prompt and the appearance and voice from an input video.
	General video editing	General video editing: Performs various video editing tasks based on input text, images, and videos. For example, it can generate a new video by extracting motion features from an input video and combining them with a prompt.
Embedding	Text embedding	Converts text into a set of numbers that represent the text. It is suitable for search, clustering, recommendation, and classification tasks.

US

In US deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are limited to the US.

Category	Subcategory	Description
Text generation	General-purpose large language models	Qwen large language models: Commercial models (Qwen-Plus, Qwen-Flash)
Text generation	Multimodal models	Visual understanding model Qwen-VL
Video generation	Text-to-video	Generates high-quality videos with rich styles from a single sentence.
Video generation	Image-to-video	First-frame-to-video: Uses an input image as the first frame and generates a video based on a prompt.
Speech recognition	Speech recognition	Qwen audio file recognition can perform speech-to-text for scenarios such as meeting transcription and live stream captioning.

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Category	Model	Description
Text generation	General-purpose large language models	Qwen large language models: Commercial models (Qwen-Max, Qwen-Plus, Qwen-Flash), open source models (Qwen3, Qwen2.5) Third-party models: DeepSeek, Kimi
	Multimodal models	Visual understanding model Qwen-VL, visual reasoning model QVQ, and omni-modal model Qwen-Omni
	Domain-specific models	Code model, Mathematical model, Translation model, Data mining model, Research model, Intention recognition model, Role-playing model
Image generation	Text-to-image	Qwen-Image: Excels in handling complex instructions, rendering both Chinese and English text, and generating high-definition, photorealistic images. It supports the selection of different models based on efficiency and quality requirements. Wan text-to-image: Basic text-to-image: Generates beautiful images from a single sentence. Mixed text and image outputs: Generates a text description followed by a corresponding image, resulting in a seamless text-and-image output. Tongyi - text-to-Image - Z-Image: A lightweight text-to-image model that quickly generates high-quality images and supports bilingual rendering in Chinese and English, complex semantic understanding, and a variety of styles and themes.
Image generation	Image editing	General-purpose models: Qwen-Image-Edit: Supports prompts in both Chinese and English, enabling complex image and text editing operations such as style transfer, text modification, and object editing. It also supports multi-image fusion and is adaptable to a wide variety of industrial application scenarios. Wan image editing: Supports scenarios such as multi-image fusion, style transfer, object detection, image restoration, and watermark removal. Model series include the following: Wan2.6, Wan2.5, Wan2.1. More models: Qwen Image Translation, OutfitAnyone
Speech synthesis and recognition	Speech synthesis (text-to-speech)	Qwen speech synthesis, Qwen realtime speech synthesis, and CosyVoice speech synthesis convert text to speech for scenarios such as voice-based customer service, audiobooks, in-car navigation, and educational tutoring.
Speech synthesis and recognition	Speech recognition and translation	Qwen realtime speech recognition, Qwen audio file recognition, Fun-ASR speech recognition, and Paraformer speech recognition convert speech to text for scenarios such as real-time meeting transcription, real-time live stream captioning, and customer service calls.
Video editing and generation	Text-to-video	Generates high-quality videos with rich styles from a single sentence.
	Image-to-video	First-frame-to-video: Generates a video from an initial image and a prompt. First-and-last-frame-to-video: Generates a video with a natural transition based on the first and last frame images and a prompt. Multi-image-to-video: Generates a video from one or more images and a text prompt, based on the entities or backgrounds in the source images. Dance video generation: AnimateAnyone generates dance videos from a character image and an action video. Image + audio to generate lip-sync videos Wan - digital human generates video from a person's image and audio. It provides a wide and natural range of motion, supports various frame sizes such as full-body, half-body, and portrait, and is suitable for scenarios such as singing and performance. EMO uses a person's image and audio to generate video with highly expressive lip-syncing and facial expressions. It supports portrait and half-body shots and is ideal for close-up scenarios. LivePortrait uses a portrait image and an audio file and is ideal for voice narration scenarios. Emoji video generation: Emoji generates facial emoji videos from facial images and preset dynamic facial templates.
	Video-to-video	Reference-to-video: Generates a video that maintains character consistency using a prompt and the appearance and voice from an input video.
	General-purpose video editing	General video editing: Performs various video editing tasks based on text prompts, images, and videos. For example, you can generate a new video by extracting motion features from an input video and combining them with a text prompt. Video lip-syncing: VideoRetalk uses a person's video and audio and is ideal for scenarios such as short video production and video translation. Video style transfer: Video style transform transforms videos into various styles, such as Japanese manga and American comics.
Vector	Text embedding	Converts text into a set of numbers that represent the text. It is used for search, clustering, recommendation, and classification.
Vector	Multimodal embedding	Converts text, images, and speech into a set of numbers. It is used for audio and video classification, image classification, and image-text retrieval.

Text generation - Qwen

The following are the Qwen commercial models. Compared to the open-source versions, the commercial models offer the latest capabilities and improvements.

The parameter sizes of the commercial models are not disclosed.

Each model is updated periodically. To use a fixed version, you can select a snapshot version. A snapshot version is typically maintained for one month after the release of the next snapshot version.

We recommend that you use the stable or latest version for more lenient rate limiting conditions.

Qwen-Max

The top-performing model in the Qwen series, suitable for complex and multi-step tasks. Usage | API reference | Try it online

Global

In the global deployment mode, endpoints and data storage are both located in the US (Virginia) region, while inference compute resources are dynamically scheduled globally.

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost	Free quota (Note)
			(Tokens)				(1,000 tokens)
qwen3-max Currently has the same capabilities as qwen3-max-2025-09-23 Context cache is available at a discount	Stable	Non-thinking only	262,144	258,048	-	65,536	Tiered pricing, see the description below the table.		None
qwen3-max-2025-09-23	Snapshot	Non-thinking only
qwen3-max-preview Context cache is available at a discount	Preview	Thinking			81,920	32,768
		Non-thinking			-	65,536

The preceding models use tiered pricing based on the number of input tokens per request.

Input tokens per request	Input price (Million tokens)	Output price (Million tokens) Chain-of-thought + response
0<Token≤32K	$1.2	$6
32K<Token≤128K	$2.4	$12
128K<Token≤252K	$3	$15

International

In the International deployment mode, endpoints and data storage are located in the Singapore region, and inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost	Free quota (Note)
			(Tokens)				(1,000 tokens)
qwen3-max Currently has the same capabilities as qwen3-max-2025-09-23 Batch calls half price	Stable	Non-thinking only	262,144	258,048	-	65,536	Tiered pricing, see the description below the table.		1 million tokens each Validity: 90 days after you activate Model Studio
qwen3-max-2025-09-23	Snapshot	Non-thinking only
qwen3-max-preview	Preview	Thinking			81,920	32,768
		Non-thinking			-	65,536

The preceding models use tiered pricing based on the number of input tokens per request.

Input tokens per request	Input price (Million tokens) qwen3-max, qwen3-max-preview support context cache.	Output price (Million tokens)
0<Token≤32K	$1.2	$6
32K<Token≤128K	$2.4	$12
128K<Token≤252K	$3	$15

More models

Model	Version	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
		(Tokens)			(Million tokens)
qwen-max Currently has the same capabilities as qwen-max-2025-01-25	Stable	32,768	30,720	8,192	$1.6 Batch calls are half price.	$6.4 Batch calls are half price.	1 million tokens each Validity: 90 days after you activate Model Studio
qwen-max-latest Always has the same capabilities as the latest snapshot version	Latest				$1.6	$6.4
qwen-max-2025-01-25 Also known as qwen-max-0125, Qwen2.5-Max	Snapshot

Mainland China

In Mainland China deployment mode, endpoints and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost
			(Tokens)				(1,000 Tokens)
qwen3-max Currently has the same capabilities as qwen3-max-2025-09-23 Batch calls are half price.	Stable	Non-thinking only	262,144	258,048	-	65,536	Tiered pricing, see the description below the table.
qwen3-max-2025-09-23	Snapshot	Non-thinking only
qwen3-max-preview	Preview	Thinking			81,920	32,768
		Non-thinking			-	65,536

The preceding models use tiered pricing based on the number of input tokens per request.

Model	Input tokens per request	Price (Million tokens)	Output price (Million tokens) Chain-of-thought + response
qwen3-max Batch calls: Half price Context cache is discounted	0<Token≤32K	$0.459	$1.836
	32K<Token≤128K	$0.918	$3.672
	128K<Token≤252K	$1.377	$5.508
qwen3-max-2025-09-23	0<Token≤32K	$0.861	$3.441
	32K<Token≤128K	$1.434	$5.735
	128K<Token≤252K	$2.151	$8.602
qwen3-max-preview Context cache is discounted	0<Token≤32K	$0.861	$3.441
	32K<Token≤128K	$1.434	$5.735
	128K<Token≤252K	$2.151	$8.602

More models

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen-max Currently has the same capabilities as qwen-max-2024-09-19	Stable	32,768	30,720	8,192	$0.345	$1.377
qwen-max-latest Always has the same capabilities as the latest snapshot version	Latest	131,072	129,024
qwen-max-2025-01-25 Also known as qwen-max-0125 or Qwen2.5-Max	Snapshot
qwen-max-2024-09-19 Also known as qwen-max-0919		32,768	30,720		$2.868	$8.602

The `thinking mode` of the qwen3-max-preview model significantly improves its overall reasoning capabilities. It performs better on agent programming, common sense reasoning, and general tasks in math and science.

Qwen-Plus

A balanced model offering a mix of performance, cost, and speed, positioned between Qwen-Max and Qwen-Flash. It is ideal for tasks of moderate complexity. Usage | API reference | Try it online｜Thinking mode

Global

In the Global deployment mode, the access point and data storage are located in the US (Virginia) region, and inference compute resources are dynamically scheduled globally.

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen-plus Currently has the same capabilities as qwen-plus-2025-07-28 Belongs to the Qwen3 series	Stable	1,000,000	Thinking mode 995,904 Non-thinking mode 997,952 The default value is 131,072, which you can change using the max_input_tokens parameter.	32,768 Max chain-of-thought is 81,920	Tiered pricing, see the description below the table.
qwen-plus-2025-12-01 Part of the Qwen3 series	Snapshot		Thinking mode 995,904 Non-thinking mode 997,952

qwen-plus-2025-09-11 Part of the Qwen3 series
qwen-plus-2025-07-28 Part of the Qwen3 series

The preceding models use tiered pricing based on the number of input tokens per request. qwen-plus supports context cache.

Input tokens per request	Input price (Million tokens)	Mode	Output price (Million tokens)
0<Token≤256K	$0.4	Non-thinking mode	$1.2
0<Token≤256K	$0.4	Thinking mode	$4
256K<Token≤1M	$1.2	Non-thinking mode	$3.6
256K<Token≤1M	$1.2	Thinking mode	$12

International

Model	Version	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
		(Tokens)			(Million tokens)
qwen-plus Currently has the same capabilities as qwen-plus-2025-07-28 Belongs to the Qwen3 series	Stable	1,000,000	Thinking mode 995,904 Non-thinking mode 997,952 The default value is 262,144. You can adjust this value with the max_input_tokens parameter.	32,768 Max chain-of-thought is 81,920	Tiered pricing, see the description below the table.		1 million tokens each Validity: 90 days after you activate Model Studio
qwen-plus-latest Currently has the same capabilities as qwen-plus-2025-12-01 Belongs to the Qwen3 series	Latest		Thinking mode 995,904 Non-thinking mode 997,952
qwen-plus-2025-12-01 Belongs to the Qwen3 series	Snapshot		Thinking mode 995,904 Non-thinking mode 997,952

qwen-plus-2025-09-11 Belongs to the Qwen3 series
qwen-plus-2025-07-28 Also known as qwen-plus-0728 Belongs to the Qwen3 series
qwen-plus-2025-07-14 Also known as qwen-plus-0714 Belongs to the Qwen3 series		131,072	Thinking mode 98,304 Non-thinking mode 129,024	16,384 Max chain-of-thought is 38,912	$0.4	Thinking mode $4 Non-thinking mode $1.2
qwen-plus-2025-04-28 Also known as qwen-plus-0428 Belongs to the Qwen3 series
qwen-plus-2025-01-25 Also known as qwen-plus-0125			129,024	8,192		$1.2

The qwen-plus, qwen-plus-latest, qwen-plus-2025-12-01, qwen-plus-2025-09-11, and qwen-plus-2025-07-28 models use tiered pricing based on the number of input tokens per request.

Input tokens per request	Input price (Million tokens)	Mode	Output price (Million tokens)
0<Token≤256K	$0.4	Non-thinking mode	$1.2
0<Token≤256K	$0.4	Thinking mode	$4
256K<Token≤1M	$1.2	Non-thinking mode	$3.6
256K<Token≤1M	$1.2	Thinking mode	$12

US

In the US deployment mode, endpoints and data storage are both located in the US (Virginia) region, and inference compute resources are limited to the United States.

Model	Version	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
		(Tokens)			(Million tokens)
qwen-plus-us Currently has the same capabilities as qwen-plus-2025-12-01-us Part of the Qwen3 series	Stable	1,000,000	Thinking mode 995,904 Non-thinking mode 997,952 The default value is 262,144. You can adjust this value with the max_input_tokens parameter.	32,768 Max chain-of-thought is 81,920	Tiered pricing, see the description below the table.		None
qwen-plus-2025-12-01-us Belongs to the Qwen3 series	Snapshot		Thinking mode 995,904 Non-thinking mode 997,952

These models use tiered pricing based on the number of input tokens per request. qwen-plus-us supports context cache.

Input tokens per request	Input price (Million tokens)	Mode	Output price (Million tokens)
0<Token≤256K	$0.4	Non-thinking mode	$1.2
0<Token≤256K	$0.4	Thinking mode	$4
256K<Token≤1M	$1.2	Non-thinking mode	$3.6
256K<Token≤1M	$1.2	Thinking mode	$12

Mainland China

In Mainland China deployment mode, the endpoints and data storage are both located in the Beijing region, while the inference compute resources are limited to Mainland China.

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen-plus Currently has the same capabilities as qwen-plus-2025-07-28 Belongs to the Qwen3 series	Stable	1,000,000	Thinking mode 995,904 Non-thinking mode 997,952 The default value is 131,072. You can adjust this value using the max_input_tokens parameter.	32,768 Max chain-of-thought is 81,920	Tiered pricing, see the description below the table.
qwen-plus-latest Currently has the same capabilities as qwen-plus-2025-12-01 Part of the Qwen3 series	Latest		Thinking mode 995,904 Non-thinking mode 997,952
qwen-plus-2025-12-01 Part of the Qwen3 series	Snapshot		Thinking mode 995,904 Non-thinking mode 997,952

qwen-plus-2025-09-11 Part of the Qwen3 series
qwen-plus-2025-07-28 Also known as qwen-plus-0728 Part of the Qwen3 series
qwen-plus-2025-07-14 Also known as qwen-plus-0714 Part of the Qwen3 series		131,072	Thinking mode 98,304 Non-thinking mode 129,024	16,384 Max chain-of-thought is 38,912	$0.115	Thinking mode $1.147 Non-thinking mode $0.287
qwen-plus-2025-04-28 Also known as qwen-plus-0428 Part of the Qwen3 series

The qwen-plus, qwen-plus-latest, qwen-plus-2025-12-01, qwen-plus-2025-09-11, and qwen-plus-2025-07-28 models use tiered pricing based on the number of input tokens per request.

Input tokens per request	Input price (Million tokens)	Mode	Output price (Million tokens)
0<Token≤128K	$0.115	Non-thinking mode	$0.287
0<Token≤128K	$0.115	Thinking mode	$1.147
128K<Token≤256K	$0.345	Non-thinking mode	$2.868
128K<Token≤256K	$0.345	Thinking mode	$3.441
256K<Token≤1M	$0.689	Non-thinking mode	$6.881
256K<Token≤1M	$0.689	Thinking mode	$9.175

The preceding models support both `thinking` and `non-thinking` modes. You can switch between the two modes using the enable_thinking parameter. In addition, the model's capabilities are significantly improved:

Reasoning: Significantly outperforms QwQ and non-reasoning models of the same size in evaluations for tasks such as math, code, and logical reasoning, reaching the top level in the industry for its scale.
Human preference: This model shows great improvements in creative writing, role assumption, multi-turn conversation, and instruction following. Its general capabilities significantly exceed those of other models of the same size.
Agent capabilities: This model achieves industry-leading performance in both `thinking` and `non-thinking` modes and can accurately call external tools.
Multilingual capabilities: Supports over 100 languages and dialects. Multilingual translation, instruction understanding, and common sense reasoning are significantly improved.
Response format: This model fixes response format issues from previous versions, such as abnormal Markdown, mid-sentence truncation, and incorrect boxed output.

For the preceding models, if `thinking mode` is enabled but no thought process is output, the `non-thinking mode` price is charged.

More models

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen-plus-2025-01-25 Also known as qwen-plus-0125	Snapshot	131,072	129,024	8,192	$0.115	$0.287
qwen-plus-2025-01-12 Also known as qwen-plus-0112
qwen-plus-2024-12-20 Also known as qwen-plus-1220
qwen-plus-2024-11-27 Also known as qwen-plus-1127
qwen-plus-2024-11-25 Also known as qwen-plus-1125
qwen-plus-2024-09-19 Also known as qwen-plus-0919
qwen-plus-2024-08-06 Also known as qwen-plus-0806			128,000		$0.574	$1.721

Qwen-Flash

The fastest and most cost-effective model in the Qwen series, ideal for simple tasks. Qwen-Flash uses flexible tiered pricing, which is more reasonable than Qwen-Turbo. Usage | API reference | Try it online | Thinking mode

Global

In the Global deployment mode, the endpoint and data storage are both located in the US (Virginia) region, and inference compute resources are dynamically scheduled globally.

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost Chain-of-thought + output
			(Tokens)				(1,000 Tokens)
qwen-flash Currently has the same capabilities as qwen-flash-2025-07-28 Part of the Qwen3 series	Stable	Thinking	1,000,000	995,904	81,920	32,768	Tiered pricing, see the description below the table.
		Non-thinking		997,952	-
qwen-flash-2025-07-28 Part of the Qwen3 series	Snapshot	Thinking		995,904	81,920
		Non-thinking		997,952	-

These models use tiered billing based on the number of tokens in the request. qwen-flash supports context cache.

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
0<Token≤256K	$0.05	$0.4
256K<Token≤1M	$0.25	$2

International

In the International deployment mode, both endpoints and data storage are located in the Singapore region, and inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost Chain-of-thought + output	Free quota (Note)
			(Tokens)				(1,000 tokens)
qwen-flash Currently has the same capabilities as qwen-flash-2025-07-28 Part of the Qwen3 series Batch calls at half price	Stable	Thinking	1,000,000	995,904	81,920	32,768	Tiered pricing, see the description below the table.		1 million tokens each Validity: 90 days after you activate Model Studio
		Non-thinking		997,952	-
qwen-flash-2025-07-28 Part of the Qwen3 series	Snapshot	Thinking		995,904	81,920
		Non-thinking		997,952	-

These models use tiered billing based on the number of tokens in each request. The qwen-flash model supports cache and Batch invocation.

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
0<Token≤256K	$0.05	$0.4
256K<Token≤1M	$0.25	$2

US

In the US deployment mode, the endpoint and data storage are both located in the US (Virginia) region, and the inference compute resource is limited to the United States.

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost Chain-of-thought + output	Free quota (Note)
			(Tokens)				(1,000 tokens)
qwen-flash-us Currently has the same capabilities as qwen-flash-2025-07-28-us Part of the Qwen3 series	Stable	Thinking	1,000,000	995,904	81,920	32,768	Tiered pricing, see the description below the table.		None
		Non-thinking		997,952	-
qwen-flash-2025-07-28-us Part of the Qwen3 series	Snapshot	Thinking		995,904	81,920
		Non-thinking		997,952	-

The preceding models use tiered pricing based on the number of input tokens per request.

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
0<Token≤256K	$0.05	$0.4
256K<Token≤1M	$0.25	$2

Mainland China

In Mainland China deployment mode, the endpoint and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost Chain-of-thought + output
			(Tokens)				Per 1,000 tokens
qwen-flash Currently has the same capabilities as qwen-flash-2025-07-28 Part of the Qwen3 series	Stable	Thinking	1,000,000	995,904	81,920	32,768	Tiered pricing, see the description below the table.
		Non-thinking		997,952	-
qwen-flash-2025-07-28 Part of the Qwen3 series	Snapshot	Thinking		995,904	81,920
		Non-thinking		997,952	-

The preceding models use tiered billing based on the number of tokens per request. The qwen-flash model supports context cache.

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
0<Token≤128K	$0.022	$0.216
128K<Token≤256K	$0.087	$0.861
256K<Token≤1M	$0.173	$1.721

Qwen-Turbo

Qwen-Turbo will no longer be updated. We recommend that you replace it with Qwen-Flash. Qwen-Flash uses flexible tiered pricing, which is more reasonable. Usage | API reference | Try it online｜Thinking mode

International

In the International deployment mode, both endpoints and data storage are located in the Singapore region, and inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model	Version	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
		(Tokens)			(Million tokens)
qwen-turbo Currently has the same capabilities as qwen-turbo-2025-04-28 Part of the Qwen3 series	Stable	Thinking mode 131,072 Non-thinking mode 1,000,000	Thinking mode 98,304 Non-thinking mode 1,000,000	16,384 Max chain-of-thought is 38,912	$0.05 Batch calls are half price.	Thinking mode: $0.5 Non-thinking mode: $0.2 Batch calls are half price.	1 million tokens each Validity: 90 days after you activate Model Studio
qwen-turbo-latest Always has the same capabilities as the latest snapshot version Part of the Qwen3 series	Latest				$0.05	Thinking mode: $0.5 Non-thinking mode: $0.2
qwen-turbo-2025-04-28 Also known as qwen-turbo-0428 Part of the Qwen3 series	Snapshot
qwen-turbo-2024-11-01 Also known as qwen-turbo-1101		1,000,000	1,000,000	8,192		$0.2

Mainland China

In Mainland China deployment mode, endpoints and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen-turbo Currently has the same capabilities as qwen-turbo-2025-04-28 Part of the Qwen3 series	Stable	Thinking mode 131,072 Non-thinking mode 1,000,000	Thinking mode 98,304 Non-thinking mode 1,000,000	16,384 Max chain-of-thought is 38,912	$0.044	Thinking mode $0.431 Non-thinking mode $0.087
qwen-turbo-latest Always has the same capabilities as the latest snapshot version Part of the Qwen3 series	Latest
qwen-turbo-2025-07-15 Also known as qwen-turbo-0715 Part of the Qwen3 series	Snapshot
qwen-turbo-2025-04-28 Also known as qwen-turbo-0428 Part of the Qwen3 series

QwQ

The QwQ reasoning model is trained based on the Qwen2.5 model. Its reasoning capabilities are significantly enhanced through reinforcement learning. Its performance on core metrics for math and code (AIME 24/25 and LiveCodeBench) and some general metrics (IFEval and LiveBench) is on par with the full-power DeepSeek-R1. Usage

International

Model

Version

Context window

Max input

Max chain-of-thought

Max response

Input cost

Output cost

Free quota

(Note)

(Tokens)

(Million tokens)

qwq-plus

Stable

131,072

98,304

32,768

8,192

$0.8

$2.4

1 million tokens

Validity: 90 days after you activate Model Studio

Mainland China

In Mainland China deployment mode, both the endpoint and data storage are located in the Beijing region, and inference compute resources are restricted to Mainland China.

Model	Version	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost
		(Tokens)				(Million tokens)
qwq-plus Currently has the same capabilities as qwq-plus-2025-03-05	Stable	131,072	98,304	32,768	8,192	$0.230	$0.574
qwq-plus-latest Always has the same capabilities as the latest snapshot version	Latest
qwq-plus-2025-03-05 Also known as qwq-plus-0305	Snapshot

Qwen-Long

The model in the Qwen series with the longest context window. It offers balanced capabilities at a low cost, making it ideal for tasks such as long-text analytics, information extraction, summarization, and classification. Usage | Try it online

Note

Only the Mainland China deployment mode is supported. In this mode, the endpoint and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen-long-latest Always has the same capabilities as the latest snapshot version	Stable	10,000,000	10,000,000	32,768	$0.072	$0.287
qwen-long-2025-01-25 Also known as qwen-long-0125	Snapshot

Qwen-Omni

The Qwen-Omni model can accept a combination of inputs in multiple modalities, such as text, images, audio, and video, and generate replies in text or speech format. It provides a variety of expressive, human-like voices and supports speech output in multiple languages and dialects. This model can be used in audio-video chat scenarios for tasks such as visual recognition, emotion analysis, education, and training. Usage｜API reference

International

In the International deployment mode, the access point and data storage are located in the Singapore region, and inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Free quota (Note)
			(Tokens)
qwen3-omni-flash Currently has the same capabilities as qwen3-omni-flash-2025-12-01	Stable	Thinking mode	65,536	16,384	32,768	16,384	1 million tokens each (regardless of modality) Validity: 90 days after you activate Model Studio
		Non-thinking mode		49,152	-
qwen3-omni-flash-2025-12-01	Snapshot	Thinking mode	65,536	16,384	32,768	16,384
		Non-thinking mode		49,152	-
qwen3-omni-flash-2025-09-15 Also known as qwen3-omni-flash-0915	Snapshot	Thinking mode	65,536	16,384	32,768	16,384
		Non-thinking mode		49,152	-

After the free quota is used, the following billing rules apply to input and output. The billing is the same for both `thinking` and `non-thinking` modes. Audio output is not supported in `thinking mode`.

Input	Price (Million tokens)
Text	$0.43
Audio	$3.81
Image/Video	$0.78

Output

Price (Million tokens)

Text

$1.66 (if input contains only text)

$3.06 (if input contains images, video, or audio)

Text + Audio

This billing item is not available in thinking mode.

$15.11 (audio)

The output text is not billed.

More models

Model	Version	Context window	Max input	Max output	Free quota (Note)
		(Tokens)
qwen-omni-turbo Same capabilities as qwen-omni-turbo-2025-03-26	Stable	32,768	30,720	2,048	1 million tokens (regardless of modality) Valid for 90 days after you activate Model Studio
qwen-omni-turbo-latest Always kept in sync with the latest snapshot version. Equivalent capabilities	Latest
qwen-omni-turbo-2025-03-26 Also known as qwen-omni-turbo-0326	Snapshot

After the free quota for the commercial model is used, the following billing rules apply to input and output:

Enter Billing Item	Price (Million tokens)
Text	$0.07
Audio input	$4.44
Image/Video	$0.21

Billable items for Outputs

Price (Million tokens)

Text

$0.27 (if the input contains only text)

$0.63 (if the input contains images, video, or audio)

Text + Audio

$8.89 (audio)

The text portion of the output is not billed.

Mainland China

In Mainland China deployment mode, both endpoints and data storage are located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Free quota (Note)
			(Tokens)
qwen3-omni-flash This model has the same capabilities as qwen3-omni-flash-2025-12-01.	Stable	Thinking mode	65,536	16,384	32,768	16,384	No free quota
		Non-thinking mode		49,152	-
qwen3-omni-flash-2025-12-01	Snapshot	Thinking mode	65,536	16,384	32,768	16,384
		Non-thinking mode		49,152	-
qwen3-omni-flash-2025-09-15 Also known as qwen3-omni-flash-0915	Snapshot	Thinking mode	65,536	16,384	32,768	16,384
		Non-thinking mode		49,152	-

Input	Price (Million tokens)
Text	$0.258
Audio	$2.265
Image/Video	$0.473

Output

Price (Million tokens)

Text

$0.989 (if the input contains only text)

$1.821 (if the input contains images, video, or audio)

Text + Audio

This billing item is not available in thinking mode.

$8.974 (audio)

The output text is not billed.

More models

Model	Version	Context window	Max input	Max output	Free quota (Note)
		Tokens
qwen-omni-turbo Has the same capabilities as qwen-omni-turbo-2025-03-26.	Stable	32,768	30,720	2,048	No free quota
qwen-omni-turbo-latest You should always use the latest snapshot version. Equivalent capabilities	Latest
qwen-omni-turbo-2025-03-26 Also known as qwen-omni-turbo-0326.	Snapshot
qwen-omni-turbo-2025-01-19 Also known as qwen-omni-turbo-0119.

The following billing rules apply to input and output:

Input	Price (Million tokens)
Text	$0.058
Audio	$3.584
Image/Video	$0.216

Output

Price (Million tokens)

Text

$0.230 (if the input is text-only)

$0.646 (if the input includes images, audio, or video)

Text + Audio

$7.168 (audio)

The output text is not billed.

Billing example: A request consists of an input with 1,000 text tokens and 1,000 image tokens, and an output with 1,000 text tokens and 1,000 audio tokens. The total cost is calculated as follows: $0.000058 (text input) + $0.000216 (image input) + $0.007168 (audio output)

We recommend using the Qwen3-Omni-Flash model first. Compared to Qwen-Omni-Turbo, which will no longer be updated, its capabilities are significantly improved:

It is a hybrid thinking model that supports both `thinking` and `non-thinking` modes. You can switch between the two modes using the enable_thinking parameter. By default, `thinking mode` is not enabled.
Audio output is not supported in `thinking mode`. In `non-thinking mode`, for the model's audio output:
- qwen3-omni-flash-2025-12-01 supports up to 49 voices, qwen3-omni-flash-2025-09-15 and qwen3-omni-flash support up to 17 voices, while Qwen-Omni-Turbo supports only 4.
- The number of supported languages is increased to 10, while Qwen-Omni-Turbo supports only 2.

Qwen-Omni-Realtime

Compared to Qwen-Omni, this model supports streaming audio input and has a built-in Voice Activity Detection (VAD) feature, which automatically detects the start and end of user speech. Usage｜Client events｜Server-side events

International

Model	Version	Context window	Max input	Max output	Free quota (Note)
		(Tokens)
qwen3-omni-flash-realtime Offers the same capabilities as qwen3-omni-flash-realtime-2025-12-01.	Stable	65,536	49,152	16,384	1 million tokens (regardless of modality) Validity: 90 days after you activate Model Studio
qwen3-omni-flash-realtime-2025-12-01	Snapshot
qwen3-omni-flash-realtime-2025-09-15

After the free quota is used, the following billing rules apply to input and output:

Input	Price (Million tokens)
Text	$0.52
Audio	$4.57
Image	$0.94

Output

Price (Million tokens)

Text

$1.99 (Text-only input)

$3.67 (Input with images or audio)

Text + Audio

$18.13 (audio)

The text output is not billed.

More models

Model	Version	Context window	Max input	Max output	Free quota (Note)
		Tokens
qwen-omni-turbo-realtime Same capabilities as qwen-omni-turbo-realtime-2025-05-08.	Stable	32,768	30,720	2,048	1 million tokens (regardless of modality) Validity: 90 days after you activate Model Studio
qwen-omni-turbo-realtime-latest Same capabilities as the latest snapshot version.	Latest
qwen-omni-turbo-realtime-2025-05-08	Snapshot

After the free quota is used, the following billing rules apply to input and output:

Input	Price (Million tokens)
Text	$0.270
Audio	$4.440
Image	$0.840

Output

Price (Million tokens)

Text

$1.070 (for text-only inputs)

$2.520 (for image or audio inputs)

Text + Audio

$8.890 (audio)

The text portion of the output is not billed.

Mainland China

In Mainland China deployment mode, endpoints and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Version	Context window	Max input	Max output	Free quota (Note)
		(Tokens)
qwen3-omni-flash-realtime This model provides the same capabilities as qwen3-omni-flash-realtime-2025-12-01.	Stable	65,536	49,152	16,384	No free quota
qwen3-omni-flash-realtime-2025-12-01	Snapshot
qwen3-omni-flash-realtime-2025-09-15

After the free quota is used, the following billing rules apply to input and output:

Input	Price (Million tokens)
Text	$0.315
Audio	$2.709
Input image	$0.559

Output

Price (Million tokens)

Text

$1.19 (for text-only input)

$2.179 (for input with images or audio)

Text + Audio

$10.766 (for audio)

The output text is not billed.

More models

Model	Version	Context window	Max input	Max output	Free quota (Note)
		(Tokens)
qwen-omni-turbo-realtime Functionally identical to qwen-omni-turbo-2025-05-08.	Stable	32,768	30,720	2,048	No free quota
qwen-omni-turbo-realtime-latest It always provides the same capabilities as the latest snapshot version.	Latest
qwen-omni-turbo-realtime-2025-05-08	Snapshot

The following billing rules apply to input and output:

Input	Price (Million tokens)
Text	$0.230
Audio	$3.584
Image	$0.861

Output

Price (Million tokens)

Text

$0.918 (for text-only input)

$2.581 (for input with images or audio)

Text + Audio

$7.168 (for the audio component)

The output text is not billed.

We recommend using the Qwen3-Omni-Flash-Realtime model first. Compared to Qwen-Omni-Turbo-Realtime, which will no longer be updated, its capabilities are significantly improved. For the model's audio output:

qwen3-omni-flash-realtime-2025-12-01 supports up to 49 voices, qwen3-omni-flash-realtime-2025-09-15 and qwen3-omni-realtime-flash support up to 17 voices, while Qwen-Omni-Turbo-Realtime supports only 4.
The number of supported languages is increased to 10, while Qwen-Omni-Turbo-Realtime supports only 2.

QVQ

QVQ is a visual reasoning model that supports visual input and chain-of-thought output. It demonstrates enhanced capabilities in math, programming, visual analysis, creation, and other general tasks. Usage | Try it online

International

In the International deployment mode, endpoints and data storage are both located in the Singapore region, and inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model	Version	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost	Free quota (Note)
		(Tokens)				(Million tokens)
qvq-max Has the same capabilities as qvq-max-2025-03-25.	Stable	131,072	106,496 A max of 16,384 tokens per image.	16,384	8,192	$1.2	$4.8	1 million tokens for each model Valid for 90 days after you activate Model Studio.
qvq-max-latest Has the same capabilities as the latest snapshot version.	Latest
qvq-max-2025-03-25 Also known as qvq-max-0325.	Snapshot

Mainland China

In Mainland China deployment mode, the endpoint and data storage are located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Version	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost
		(Tokens)				(Million tokens)
qvq-max Compared to qvq-plus, this model provides superior visual reasoning and instruction-following capabilities and is optimized for complex tasks. This model currently provides the same capabilities as qvq-max-2025-03-25.	Stable	131,072	106,496 A max of 16,384 per image.	16,384	8,192	$1.147	$4.588
qvq-max-latest This model always provides the same capabilities as the latest snapshot version.	Latest
qvq-max-2025-05-15 Also known as qvq-max-0515.	Snapshot
qvq-max-2025-03-25 Also known as qvq-max-0325.
qvq-plus This model currently provides the same capabilities as qvq-plus-2025-05-15.	Stable					$0.287	$0.717
qvq-plus-latest This model always provides the same capabilities as the latest snapshot version.	Latest
qvq-plus-2025-05-15 Also known as qvq-plus-0515.	Snapshot

Qwen-VL

Qwen-VL is a text generation model with visual (image) understanding capabilities. It can perform Optical Character Recognition (OCR), summarization, and reasoning. For example, it can extract attributes from product photos or solve problems based on diagrams in exercises. Usage | API reference | Try it online

The Qwen-VL model is billed based on the total number of input and output tokens. For more information about the calculation rules for image tokens, see Visual understanding.

Global

In Global deployment mode, endpoints and data storage are both located in the US (Virginia) region, and inference compute resources are dynamically scheduled globally.

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost Chain-of-thought and output
			(Tokens)				(Million tokens)
qwen3-vl-plus Currently provides the same capabilities as qwen3-vl-plus-2025-09-23	Stable	Thinking	262,144	258,048 Max of 16,384 tokens per image	81,920	32,768	Tiered pricing applies. For more information, see the description below the table.
		Non-thinking		260,096 Max of 16,384 tokens per image	-
qwen3-vl-plus-2025-09-23	Snapshot	Thinking		258,048 A single graph supports a max of 16,384.	81,920
		Non-thinking		260,096 Max per graph: 16,384	-
qwen3-vl-flash Currently provides the same capabilities as qwen3-vl-flash-2025-10-15	Stable	Thinking		258,048 Max of 16,384 tokens per image	81,920
		Non-thinking		260,096 A single graph has a max capacity of 16,384.	-
qwen3-vl-flash-2025-10-15	Snapshot	Thinking		258,048 Max of 16,384 tokens per image	81,920
		Non-thinking		260,096 Max of 16,384 tokens per image	-

The preceding models use tiered billing based on the number of input tokens for each request. The input and output prices are the same for both `thinking` and `non-thinking` modes.

qwen3-vl-plus series

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
0 < Tokens ≤ 32K	$0.2	$1.6
32K < Tokens ≤ 128K	$0.3	$2.4
128K < Tokens ≤ 256K	$0.6	$4.8

qwen3-vl-flash series

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
Up to 32K	$0.05	$0.4
Over 32K to 128K	$0.075	$0.6
Over 128K to 256K	$0.12	$0.96

International

In International deployment mode, the endpoint and data storage are located in the Singapore region, and inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost Chain-of-thought and output	Free quota (Note)
			(Tokens)				(Million tokens)
qwen3-vl-plus Offers the same capabilities as qwen3-vl-plus-2025-09-23	Stable	Thinking	262,144	258,048 Max 16,384 per image	81,920	32,768	Tiered pricing, see the description below the table.		Each contains 1 million tokens. Validity: 90 days after you activate Model Studio
		Non-thinking		260,096 Max 16,384 per image	-
qwen3-vl-plus-2025-12-19	Snapshot	Thinking		258,048 Max 16,384 per image	81,920
		Non-thinking		260,096 Max 16,384 per image	-
qwen3-vl-plus-2025-09-23	Snapshot	Thinking		258,048 Max 16,384 per image	81,920
		Non-thinking		260,096 Max 16,384 per image	-
qwen3-vl-flash Offers the same capabilities as qwen3-vl-flash-2025-10-15	Stable	Thinking		258,048 Max 16,384 per image	81,920
		Non-thinking		260,096 Max 16,384 per image	-
qwen3-vl-flash-2025-10-15	Snapshot	Thinking		258,048 Max 16,384 per image	81,920
		Non-thinking		260,096 Max 16,384 per image	-

The preceding models use tiered pricing based on the number of input tokens per request. The input and output prices are the same for both `thinking` and `non-thinking` modes.

qwen3-vl-plus series

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
0 < Tokens ≤ 32K	$0.20	$1.60
32K < Tokens ≤ 128K	$0.30	2.4
128K < Tokens ≤ 256K	$0.60	$4.80

qwen3-vl-flash series

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
0 < Tokens ≤ 32K	$0.05	$0.4
32K < Tokens ≤ 128K	$0.075	$0.6
128K < Tokens ≤ 256K	$0.12	$0.96

More models

Qwen-VL-Max

Qwen-VL-Max performs better than Qwen-VL-Plus. The following models are all part of the Qwen2.5-VL series.

Model	Version	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
		(Tokens)			(Million tokens)
qwen-vl-max This model is an improvement on qwen-vl-plus. It features enhanced visual reasoning and instruction-following capabilities and delivers optimal performance for complex tasks. Its capabilities are identical to those of qwen-vl-max-2025-08-13.	Stable	131,072	129,024 A max of 16,384 tokens per image.	8,192	$0.8 Batch calls are half price.	$3.2 Batch calls are half price.	1 million tokens each Valid for 90 days after you activate Model Studio
qwen-vl-max-latest This model always has the same capabilities as the latest snapshot version.	Latest				$0.8	$3.2
qwen-vl-max-2025-08-13 Also known as qwen-vl-max-0813. This version features comprehensive improvements to visual understanding metrics, with significant enhancements in areas such as math, reasoning, object recognition, and multilingual processing.	Snapshot
qwen-vl-max-2025-04-08 Also known as qwen-vl-max-0408. As part of the Qwen2.5-VL model series, this model extends the context window to 128,000 tokens and offers significantly enhanced mathematical and reasoning capabilities.

Qwen-VL-Plus

The Qwen-VL-Plus model offers a balance of performance and cost. The following models are all part of the Qwen2.5-VL series.

Model	Version	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
		(Tokens)			(Million tokens)
qwen-vl-plus Functionally identical to qwen-vl-plus-2025-08-15.	Stable	131,072	129,024 A max of 16,384 tokens per image.	8,192	$0.21 Batch calls cost 50% less.	$0.63 Batch calls cost 50% less.	1 million tokens each Valid for 90 days after you activate Model Studio.
qwen-vl-plus-latest Functionally identical to the latest snapshot version.	Latest				$0.21	$0.63
qwen-vl-plus-2025-08-15 Also known as qwen-vl-plus-0815. Significant improvements in object recognition, localization, and multilingual processing.	Snapshot
qwen-vl-plus-2025-05-07 Also known as qwen-vl-plus-0507. Significantly improved understanding of math, reasoning, and surveillance video content.
qwen-vl-plus-2025-01-25 Also known as qwen-vl-plus-0125. As a part of the Qwen2.5-VL model series, this model extends the context window to 128k tokens and significantly enhances image and video understanding.

US

In the US deployment mode, the endpoint and data storage are located in the US (Virginia) region, and the inference compute resource is limited to the United States.

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost Chain-of-thought and output
			(Tokens)				(Million tokens)
qwen3-vl-flash-us Same capabilities as qwen3-vl-flash-2025-10-15-us	Stable	Thinking		258,048 Max 16,384 per image	81,920	32,768	Tiered pricing, see the description below the table.
		Non-thinking		260,096 Max 16,384 per image	-
qwen3-vl-flash-2025-10-15-us	Snapshot	Thinking		258,048 Max 16,384 per image	81,920
		Non-thinking		260,096 Max 16,384 per image	-

The preceding models are billed on a tiered basis according to the number of input tokens in the request. The input and output prices are the same for both thinking mode and non-thinking mode. The qwen3-vl-flash-us model supports context cache.

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
Up to 32K	$0.05	$0.4
Over 32K, up to 128K	$0.075	$0.6
Over 128K, up to 256K	$0.12	$0.96

Mainland China

In Mainland China deployment mode, the endpoint and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Version	Mode	Context window	Max input	Max chain-of-thought	Max output	Input cost	Output cost	Free quota (Note)
			(Tokens)				(Million tokens)
qwen3-vl-plus Has the same capabilities as qwen3-vl-plus-2025-09-23.	Stable	Thinking	262,144	258,048 A max of 16,384 per image	81,920	32,768	Tiered pricing applies. For details, see the description below this table.		No free quota
		Non-thinking		260,096 A max of 16,384 per image	-
qwen3-vl-plus-2025-12-19	Snapshot	Thinking		258,048 A max of 16,384 per image	81,920
		Non-thinking		260,096 A max of 16,384 per image	-
qwen3-vl-plus-2025-09-23	Snapshot	Thinking		258,048 A max of 16,384 per image	81,920
		Non-thinking		260,096 A max of 16,384 per image	-
qwen3-vl-flash Has the same capabilities as qwen3-vl-flash-2025-10-15.	Stable	Thinking		258,048 A max of 16,384 per image	81,920
		Non-thinking		260,096 A max of 16,384 per image	-
qwen3-vl-flash-2025-10-15	Snapshot	Thinking		258,048 A max of 16,384 per image	81,920
		Non-thinking		260,096 A max of 16,384 per image	-

The preceding models use tiered pricing based on the number of input tokens per request. The input and output prices are the same for both `thinking` and `non-thinking` modes.

qwen3-vl-plus series

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
0 < tokens ≤ 32K	$0.143	$1.434
32K < tokens ≤ 128K	$0.215	$2.15
128K < tokens ≤ 256K	$0.43	$4.301

qwen3-vl-flash series

Input tokens per request	Input price (Million tokens)	Output price (Million tokens)
0 < Tokens ≤ 32K	$0.022	$0.215
32K < Tokens ≤ 128K	$0.043	$0.43
128K < Tokens ≤ 256K	$0.086	$0.859

More models

Qwen-VL-Max series

Models updated on or after qwen-vl-max-2025-01-25 are part of the Qwen2.5-VL series.

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(in tokens)			(cost Million tokens)
qwen-vl-max Offers improved visual reasoning and instruction-following capabilities compared to qwen-vl-plus, and provides optimal performance for more complex tasks. Currently has the same capabilities as qwen-vl-max-2025-08-13.	Stable	131,072	129,024 Max of 16,384 per image	8,192	$0.23	$0.574
qwen-vl-max-latest Always equivalent to the latest snapshot version.	Latest
qwen-vl-max-2025-08-13 Also known as qwen-vl-max-0813. Comprehensively improves visual understanding, with significant enhancements in math, reasoning, object recognition, and multilingual processing.	Snapshot
qwen-vl-max-2025-04-08 Also known as qwen-vl-max-0408. Enhances math and reasoning capabilities.					$0.431	$1.291
qwen-vl-max-2025-04-02 Also known as qwen-vl-max-0402. Significantly improves the accuracy of solving complex math problems.
qwen-vl-max-2025-01-25 Also known as qwen-vl-max-0125. This version is an upgrade to the Qwen2.5-VL series. It features an extended context window of 128,000 tokens and has significantly enhanced image and video understanding capabilities.
qwen-vl-max-2024-12-30 Also known as qwen-vl-max-1230.		32,768	30,720 Max of 16,384 per image	2,048	$0.431	$1.291
qwen-vl-max-2024-11-19 Also known as qwen-vl-max-1119.
qwen-vl-max-2024-10-30 Also known as qwen-vl-max-1030.					$2.868
qwen-vl-max-2024-08-09 Also known as qwen-vl-max-0809.

Qwen-VL-Plus series

Models updated on or after qwen-vl-plus-2025-01-25 are part of the Qwen2.5-VL series.

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen-vl-plus Provides the same capabilities as qwen-vl-plus-2025-08-15.	Stable	131,072	129,024 Max of 16,384 per image.	8,192	$0.115	$0.287
qwen-vl-plus-latest Mirrors the capabilities of the latest snapshot version.	Latest
qwen-vl-plus-2025-08-15 Also known as qwen-vl-plus-0815. Provides significant improvements in object recognition, localization, and multilingual processing.	Snapshot
qwen-vl-plus-2025-07-10 Also known as qwen-vl-plus-0710. Provides improved understanding of surveillance video content.		32,768	30,720 Max of 16,384 per image.		$0.022	$0.216
qwen-vl-plus-2025-05-07 Also known as qwen-vl-plus-0507. Provides significantly improved understanding of math, reasoning, and surveillance video content.		131,072	129,024 Max of 16,384 per image.		$0.216	$0.646
qwen-vl-plus-2025-01-25 Also known as qwen-vl-plus-0125. Upgraded to the Qwen2.5-VL series. This version extends the context to 128k and significantly enhances image and video understanding.
qwen-vl-plus-2025-01-02 Also known as qwen-vl-plus-0102.		32,768	30,720 Max of 16,384 per image.	2,048
qwen-vl-plus-2024-08-09 Also known as qwen-vl-plus-0809.

Qwen-OCR

The Qwen-OCR model specializes in text extraction. Compared to the Qwen-VL model, Qwen-OCR focuses more on extracting text from images of documents, tables, exam questions, and handwriting. It can recognize multiple languages, including English, French, Japanese, Korean, German, Russian, and Italian. Usage | API reference｜Try it online

Global

In the global deployment mode, both the endpoint and data storage are located in the US (Virginia) region, and inference compute resources are dynamically scheduled globally.

Model	Version	Context window	Max input	Max output	Input price	Output price
		(Tokens)			(Million tokens)
qwen-vl-ocr Provides the same capabilities as qwen-vl-ocr-2025-11-20.	Stable	34,096	30,000 Max of 30,000 per image.	4096	$0.07	$0.16
qwen-vl-ocr-2025-11-20 Also known as qwen-vl-ocr-1120. This model is based on the Qwen3-VL architecture and provides significantly improved document parsing and text localization capabilities.	Snapshot	38,192		8,192

International

Model	Version	Context window	Max input	Max output	Input price	Output price	Free quota (Note)
Model	Version	(Tokens)			(Million tokens)		Free quota (Note)
qwen-vl-ocr	Stable	34,096	30,000 Max of 30,000 per image.	4,096	$0.72	$0.72	1 million input tokens and 1 million output tokens Valid for 90 days after you activate Model Studio.
qwen-vl-ocr-2025-11-20 Also known as qwen-vl-ocr-1120. Based on the Qwen3-VL architecture, this model offers significantly improved document parsing and text localization.	Snapshot	38,192	30,000 Max of 30,000 per image.	8,192	$0.07	$0.16

Mainland China

In Mainland China deployment mode, endpoints and data storage are located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Version	Context window	Max input	Max output	Input price	Output price	Free quota (Note)
		(Tokens)			(Million tokens)
qwen-vl-ocr Equivalent to qwen-vl-ocr-2025-08-28.	Stable	34,096	30,000 Max 30,000 per image.	4,096	$0.717	$0.717	No free quota
qwen-vl-ocr-latest It always has the capabilities of the latest version.	Latest	38,192		8,192	$0.043	$0.072
qwen-vl-ocr-2025-11-20 Also known as qwen-vl-ocr-1120. Based on the Qwen3-VL architecture, this version provides significantly improved document parsing and text localization capabilities.	Snapshot
qwen-vl-ocr-2025-08-28 Also known as qwen-vl-ocr-0828.		34,096		4,096	$0.717	$0.717
qwen-vl-ocr-2025-04-13 Also known as qwen-vl-ocr-0413.
qwen-vl-ocr-2024-10-28 Also known as qwen-vl-ocr-1028.

Qwen-Math

Qwen-Math is a language model that specializes in solving math problems. Usage | API reference | Try it online

Note

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen-math-plus Equivalent to qwen-math-plus-2024-09-19	Stable	4,096	3,072	3,072	$0.574	$1.721
qwen-math-plus-latest Equivalent to the latest snapshot version	Latest
qwen-math-plus-2024-09-19 Also known as qwen-math-plus-0919	Snapshot
qwen-math-plus-2024-08-16 Also known as qwen-math-plus-0816
qwen-math-turbo Equivalent to qwen-math-turbo-2024-09-19	Stable				$0.287	$0.861
qwen-math-turbo-latest Equivalent to the latest snapshot version	Latest
qwen-math-turbo-2024-09-19 Also known as qwen-math-turbo-0919	Snapshot

Qwen-Coder

Qwen-Coder is a code model. The latest Qwen3-Coder-Plus series is a code generation model based on Qwen3. It has powerful coding agent capabilities, excels at tool calling and environment interaction, and can perform autonomous programming. It has excellent coding and general-purpose capabilities. Usage | API reference | Try it online

Global

In the Global deployment mode, endpoints and data storage are located in the US (Virginia) region, while inference compute resources are dynamically scheduled globally.

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen3-coder-plus Currently identical to qwen3-coder-plus-2025-09-23.	Stable	1,000,000	997,952	65,536	Tiered pricing applies. For more information, see the description below this table.
qwen3-coder-plus-2025-09-23	Snapshot
qwen3-coder-plus-2025-07-22	Snapshot
qwen3-coder-flash Currently identical to qwen3-coder-flash-2025-07-28.	Stable
qwen3-coder-flash-2025-07-28	Snapshot

The preceding models use tiered pricing based on the number of input tokens per request.

qwen3-coder-plus series

The prices for qwen3-coder-plus, qwen3-coder-plus-2025-09-23, and qwen3-coder-plus-2025-07-22 are as follows. qwen3-coder-plus supports context cache. Input text that hits the implicit cache is billed at 20% of the price.

Input tokens per request	Input cost (Million tokens)	Output cost (Million tokens)
0 < Tokens ≤ 32K	$1	$5
32K < Tokens ≤ 128K	$1.8	$9
128K < Tokens ≤ 256K	$3	$15
256K < Tokens ≤ 1M	$6	$60

qwen3-coder-flash series

The prices for qwen3-coder-flash and qwen3-coder-flash-2025-07-28 are as follows. qwen3-coder-flash supports context cache. Input text that hits the cache is billed at 20% of the price.

Input tokens per request	Input cost (Million tokens)	Output cost (Million tokens)
0 < Tokens ≤ 32K	$0.3	$1.5
32K < Tokens ≤ 128K	$0.5	$2.5
128K < Tokens ≤ 256K	$0.8	$4
256K < Tokens ≤ 1M	$1.6	$9.6

International

Model	Version	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
		(Tokens)			(Million tokens)
qwen3-coder-plus Same capabilities as qwen3-coder-plus-2025-09-23	Stable	1,000,000	997,952	65,536	Tiered pricing, see the description below the table.		1 million tokens each Validity: 90 days after you activate Model Studio
qwen3-coder-plus-2025-09-23	Snapshot
qwen3-coder-plus-2025-07-22	Snapshot
qwen3-coder-flash Same capabilities as qwen3-coder-flash-2025-07-28	Stable
qwen3-coder-flash-2025-07-28	Snapshot

The preceding models use tiered pricing based on the number of input tokens per request.

qwen3-coder-plus series

Input tokens per request	Input cost (Million tokens)	Output cost (Million tokens)
0 < tokens ≤ 32K	$1	$5
32K < tokens ≤ 128K	$1.8	$9
128K < tokens ≤ 256K	$3	$15
256K < tokens ≤ 1M	$6	$60

qwen3-coder-flash series

The prices for qwen3-coder-flash and qwen3-coder-flash-2025-07-28 are as follows. qwen3-coder-flash supports context cache. Input text that hits the implicit cache is billed at 20% of the price, and input text that hits the explicit cache is billed at 10% of the price.

Input tokens per request	Input cost (Million tokens)	Output cost (Million tokens)
0 < Tokens ≤ 32K	$0.30	$1.50
32K < Tokens ≤ 128K	$0.50	$2.50
128K < Tokens ≤ 256K	$0.80	$4.00
256K < Tokens ≤ 1M	$1.6	$9.6

Mainland China

In Mainland China deployment mode, endpoints and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million Tokens)
qwen3-coder-plus This is functionally equivalent to qwen3-coder-plus-2025-09-23.	Stable	1,000,000	997,952	65,536	Tiered pricing applies. See the notes below the table.
qwen3-coder-plus-2025-09-23	Snapshot
qwen3-coder-plus-2025-07-22	Snapshot
qwen3-coder-flash The current capabilities are the same as those of qwen3-coder-flash-2025-07-28.	Stable
qwen3-coder-flash-2025-07-28	Snapshot

The preceding models use tiered pricing based on the number of input tokens per request.

qwen3-coder-plus series

Input tokens per request	Input cost (Million tokens)	Output cost (Million tokens)
0 < Tokens ≤ 32K	$0.574	$2.294
32K < Tokens ≤ 128K	$0.861	$3.441
128K < Tokens ≤ 256K	$1.434	$5.735
256K < Tokens ≤ 1M	$2.868	$28.671

qwen3-coder-flash series

Input tokens per request	Input cost (Million tokens)	Output cost (Million tokens)
Up to 32 K	$0.144	$0.574
Over 32 K to 128 K	$0.216	$0.861
Over 128 K to 256 K	$0.359	$1.434
Over 256 K to 1 M	$0.717	$3.584

More models

Model	Version	Context window	Max input	Max output	Input cost	Output cost
		(Tokens)			(Million tokens)
qwen-coder-plus Equivalent to qwen-coder-plus-2024-11-06	Stable	131,072	129,024	8,192	$0.502	$1.004
qwen-coder-plus-latest Equivalent to the latest snapshot version of qwen-coder-plus	Latest
qwen-coder-plus-2024-11-06 Also known as qwen-coder-plus-1106	Snapshot
qwen-coder-turbo Equivalent to qwen-coder-turbo-2024-09-19	Stable	131,072	129,024	8,192	$0.287	$0.861
qwen-coder-turbo-latest Equivalent to the latest snapshot version of qwen-coder-turbo	Latest
qwen-coder-turbo-2024-09-19 Also known as qwen-coder-turbo-0919	Snapshot

Qwen-MT

Qwen-MT is a flagship large translation model that is fully upgraded based on Qwen 3. It supports mutual translation between 92 languages, including Chinese, English, Japanese, Korean, French, Spanish, German, Thai, Indonesian, Vietnamese, and Arabic. The model's performance and translation quality are fully upgraded. It provides more stable term customization, format restoration, and domain prompt capabilities, which makes translations more accurate and natural. Usage

Global

In the global deployment mode, the endpoint and data storage are both located in the US (Virginia) region, while inference compute resources are dynamically scheduled globally.

Model	Context window	Max input	Max output	Input cost	Output cost
	(Tokens)			(Million tokens)
qwen-mt-plus Part of the Qwen3-MT	16,384	8,192	8,192	$2.46	$7.37
qwen-mt-flash Part of the Qwen3-MT				$0.16	$0.49
qwen-mt-lite Part of the Qwen3-MT				$0.12	$0.36

International

In the International deployment mode, the endpoint and data storage are both located in the Singapore region, and inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model	Context window	Max input	Max output	Input cost	Output cost	Free quota Rule description
	(Tokens)			(Million tokens)
qwen-mt-plus Part of the Qwen3-MT	16,384	8,192	8,192	$2.46	$7.37	1 million tokens Valid for 90 days after you activate Model Studio.
qwen-mt-flash Part of the Qwen3-MT				$0.16	$0.49
qwen-mt-lite Part of the Qwen3-MT				$0.12	$0.36
qwen-mt-turbo Part of the Qwen3-MT				$0.16	$0.49

Mainland China

In Mainland China deployment mode, endpoints and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Context window	Max input	Max output	Input cost	Output cost
Model	(tokens)			(Million tokens)
qwen-mt-plus Part of the Qwen3-MT	16,384	8,192	8,192	$0.259	$0.775
qwen-mt-flash Part of the Qwen3-MT				$0.101	$0.280
qwen-mt-lite Part of the Qwen3-MT				$0.086	$0.229
qwen-mt-turbo Part of the Qwen3-MT				$0.101	$0.280

Qwen data mining model

The Qwen data mining model can extract structured information from documents and can be used in fields such as data annotation and content moderation. Usage | API reference

Note

Model	Context window	Max input	Max output	Input cost	Output cost	Free quota
Model	(Tokens)			(Million tokens)		Free quota
qwen-doc-turbo	262,144	253,952	32,768	$0.087	$0.144	No free quota

Qwen deep research model

The Qwen deep research model can break down complex problems, conduct reasoning and analysis in combination with internet searches, and generate research reports. Usage | API reference

Note

Model	Context window	Max input	Max output	Input cost	Output cost
Model	(Tokens)			(1,000 tokens)
qwen-deep-research	1,000,000	997,952	32,768	$0.007742	$0.023367

Text generation - Qwen - Open source

In the model names, xxb indicates the parameter size. For example, qwen2-72b-instruct indicates a parameter size of 72 billion (72B).
Model Studio supports invoking the open-source versions of Qwen. You do not need to deploy the models locally. For open-source versions, we recommend using the Qwen3 and Qwen2.5 models.

Qwen3

The qwen3-next-80b-a3b-thinking model, released in September 2025, supports only `thinking mode`. It offers improved instruction-following capabilities and more concise summaries compared to qwen3-235b-a22b-thinking-2507.

The qwen3-next-80b-a3b-instruct model, released in September 2025, supports only `non-thinking mode`. It provides enhanced Chinese comprehension, logical reasoning, and text generation capabilities compared to qwen3-235b-a22b-instruct-2507.

The qwen3-235b-a22b-thinking-2507 and qwen3-30b-a3b-thinking-2507 models, released in July 2025, support only `thinking mode`. They are upgraded versions of qwen3-235b-a22b (`thinking mode`) and qwen3-30b-a3b (`thinking mode`).

The qwen3-235b-a22b-instruct-2507 and qwen3-30b-a3b-instruct-2507 models, released in July 2025, support only `non-thinking mode`. They are upgraded versions of qwen3-235b-a22b (`non-thinking mode`) and qwen3-30b-a3b (`non-thinking mode`).

The Qwen3 models, released in April 2025, support both `thinking` and `non-thinking` modes. You can use the enable_thinking parameter to switch between the modes. The Qwen3 models also feature significant capability enhancements:

Reasoning capabilities: In evaluations for math, code, and logical reasoning, these models significantly outperform QwQ and non-reasoning models of the same size, reaching top-tier industry performance for their scale.
Human preference alignment: These models show greatly improved capabilities in creative writing, role-playing, multi-turn conversation, and instruction following. Their general capabilities significantly surpass those of other models of the same size.
Agent capabilities: These models reach industry-leading levels in both `thinking` and `non-thinking` modes, enabling precise external tool calling.

Multilingual capabilities: Supports over 100 languages and dialects, with significant improvements in multilingual translation, instruction understanding, and common-sense reasoning.

Supported languages

English

Simplified Chinese

Traditional Chinese

French

Spanish

Arabic is written in the Arabic script and is the official language of many Arab countries.

Russian is written in the Cyrillic script and is the official language of Russia and some other countries.

Portuguese is written in the Latin script and is the official language of Portugal, Brazil, and other Portuguese-speaking countries.

German is written in the Latin script and is the official language in countries such as Germany and Austria.

Italian is written in the Latin script and is the official language of Italy, San Marino, and parts of Switzerland.

Dutch is written in the Latin script and is the official language of the Netherlands, parts of Belgium (Flanders), and Suriname.

Danish is written in the Latin script and is the official language of Denmark.

Irish is written in the Latin script and is one of the official languages of Ireland.

Welsh is written in the Latin script. It is spoken in Wales and is one of its official languages.

Finnish is written in the Latin script and is the official language of Finland.

Icelandic is written in the Latin script and is the official language of Iceland.

Swedish is written in the Latin script and is the official language of Sweden.

Norwegian Nynorsk is written in the Latin script and is one of the two official written standards for the Norwegian language.

Norwegian Bokmål is written in the Latin script and is the more common of the two official written standards for the Norwegian language.

Japanese is written in the Japanese script and is the official language of Japan.

Korean is written in Hangul and is the official language of South Korea and North Korea.

Vietnamese is written in the Latin script and is the official language of Vietnam.

Thai is written in the Thai script and is the official language of Thailand.

Indonesian is written in the Latin script and is the official language of Indonesia.

Malay is written in the Latin script and is a major language in countries such as Malaysia.

Burmese is written in the Burmese script and is the official language of Myanmar.

Tagalog is written in the Latin script and is one of the major languages of the Philippines.

Khmer is written in the Khmer script and is the official language of Cambodia.

Lao is written in the Lao script and is the official language of Laos.

Hindi is written in Devanagari and is one of the official languages of India.

Bengali is written in the Bengali script and is the official language of Bangladesh and the Indian state of West Bengal.

Urdu is written in the Arabic script. It is an official language of Pakistan and is also widely spoken in India.

Nepali is written in the Devanagari script and is the official language of Nepal.

Hebrew is written in the Hebrew script and is the official language of Israel.

Turkish is written in the Latin script and is the official language of Türkiye and Northern Cyprus.

Persian is written in the Arabic script and is the official language in countries such as Iran and Tajikistan.

Polish is written in the Latin script and is the official language of Poland.

Ukrainian is written in the Cyrillic script and is the official language of Ukraine.

Czech is written in the Latin script and is the official language of the Czech Republic.

Romanian is written in the Latin script and is the official language of Romania and Moldova.

Bulgarian is written in the Cyrillic script and is the official language of Bulgaria.

Slovak is written in the Latin script and is the official language of Slovakia.

Hungarian is written in the Latin script and is the official language of Hungary.

Slovenian is written in the Latin script and is the official language of Slovenia.

Latvian is written in the Latin script and is the official language of Latvia.

Estonian is written in the Latin script and is the official language of Estonia.

Lithuanian is written in the Latin script and is the official language of Lithuania.

Belarusian is written in the Cyrillic script and is one of the official languages of Belarus.

Greek is written in the Greek alphabet and is the official language of Greece and Cyprus.

Croatian is written in the Latin script and is the official language of Croatia.

Macedonian is written in the Cyrillic script and is the official language of North Macedonia.

Maltese is written in the Latin script and is the official language of Malta.

Serbian is written in the Cyrillic script and is the official language of Serbia.

Bosnian is written in the Latin script and is one of the official languages of Bosnia and Herzegovina.

Georgian is written in the Georgian script and is the official language of Georgia.

Armenian is written in the Armenian alphabet and is the official language of Armenia.

North Azerbaijani is written in the Latin script and is the official language of Azerbaijan.

Kazakh is written in the Cyrillic script and is the official language of Kazakhstan.

Northern Uzbek is written in the Latin script and is the official language of Uzbekistan.

Tajik is written in the Cyrillic script and is the official language of Tajikistan.

Swahili is written in the Latin script. It is a lingua franca or official language in many East African countries.

Afrikaans is written in the Latin script and is mainly spoken in South Africa and Namibia.

Cantonese is written in Traditional Chinese characters. It is mainly spoken in Guangdong Province, Hong Kong, and Macao, and is a primary language in these regions.

Luxembourgish is written in the Latin script. It is spoken in Luxembourg and parts of Germany, and is one of the country's official languages.

Limburgish is written in the Latin script and is mainly spoken in the Netherlands, Belgium, and parts of Germany.

Catalan is written in the Latin script and is spoken in Catalonia and other parts of Spain.

Galician is written in the Latin script and is mainly spoken in Galicia, Spain.

Asturian is written in the Latin script and is mainly spoken in Asturias, Spain.

Basque is written in the Latin script. It is mainly spoken in the Basque Country of Spain and France, and is an official language of the Basque Autonomous Community in Spain.

Occitan is written in the Latin script and is mainly spoken in Southern France.

Venetian is written in the Latin script and is mainly spoken in the Veneto region of Italy.

Sardinian is written in the Latin script and is mainly spoken on the island of Sardinia in Italy.

Sicilian is written in the Latin script and is mainly spoken on the island of Sicily in Italy.

Friulian is written in the Latin script and is mainly spoken in the Friuli-Venezia Giulia region of Italy.

Lombard is written in the Latin script and is mainly spoken in the Lombardy region of Italy.

Ligurian is written in the Latin script and is mainly spoken in the Liguria region of Italy.

Faroese is written in the Latin script. It is mainly spoken in the Faroe Islands and is one of their official languages.

Tosk Albanian is written in the Latin script and is the southern dialect of Albanian.

Silesian is written in the Latin script and is mainly spoken in Poland.

Bashkir is written in the Cyrillic script and is mainly spoken in Bashkortostan, Russia.

Tatar is written in the Cyrillic script and is mainly spoken in Tatarstan, Russia.

Mesopotamian Arabic is written in the Arabic script and is mainly spoken in Iraq.

Najdi Arabic is written in the Arabic script and is mainly spoken in the Najd region of Saudi Arabia.

Egyptian Arabic is written in the Arabic script and is mainly spoken in Egypt.

Levantine Arabic is written in the Arabic script and is mainly spoken in Syria and Lebanon.

Ta'izzi-Adeni Arabic is written in the Arabic script and is mainly spoken in southern Yemen.

Dari is written in the Arabic script and is one of the official languages of Afghanistan.

Tunisian Arabic is written in the Arabic script and is mainly spoken in Tunisia.

Moroccan Arabic is written in the Arabic script and is mainly spoken in Morocco.

Kabuverdianu is written in the Latin script and is mainly spoken in Cape Verde.

Tok Pisin uses the Latin alphabet and is one of the main lingua francas in Papua New Guinea.

Eastern Yiddish is written in the Hebrew script and is mainly spoken in Jewish communities.

Sindhi is written in the Arabic script and is an official language of the Sindh province in Pakistan.

Sinhala is written in the Sinhala script and is one of the official languages of Sri Lanka.

Telugu is written in the Telugu script and is an official language in the Indian states of Andhra Pradesh and Telangana.

Punjabi is written in the Gurmukhi script. It is spoken in the Indian state of Punjab and is one of India's official languages.

Tamil is written in the Tamil script and is an official language in the Indian state of Tamil Nadu and in Sri Lanka.

Gujarati is written in the Gujarati script and is an official language of the Indian state of Gujarat.

Malayalam is written in the Malayalam script and is an official language of the Indian state of Kerala.

Marathi is written in the Devanagari script and is an official language of the Indian state of Maharashtra.

Kannada is written in the Kannada script and is an official language of the Indian state of Karnataka.

Magahi is written in the Devanagari script and is mainly spoken in the Indian state of Bihar.

Oriya is written in the Odia script and is an official language of the Indian state of Odisha.

Awadhi is written in the Devanagari script and is mainly spoken in the Indian state of Uttar Pradesh.

Maithili is written in the Devanagari script. It is spoken in the Indian state of Bihar and the Terai plains of Nepal, and is one of the official languages of India.

Assamese is written in the Bengali script and is an official language of the Indian state of Assam.

Chhattisgarhi is written in the Devanagari script and is mainly spoken in the Indian state of Chhattisgarh.

Bhojpuri is written in the Devanagari script and is spoken in parts of India and Nepal.

Minangkabau is written in the Latin script and is mainly spoken on the island of Sumatra in Indonesia.

Balinese is written in the Latin script and is mainly spoken on the island of Bali in Indonesia.

Javanese is written in the Latin script, although the Javanese script is also commonly used. It is widely spoken on the island of Java in Indonesia.

Banjar is written in the Latin script and is mainly spoken on the island of Kalimantan in Indonesia.

Sundanese is written in the Latin script, although the Sundanese script is also traditionally used. It is mainly spoken in the western part of the island of Java in Indonesia.

Cebuano is written in the Latin script and is mainly spoken in the Central Visayas and parts of Mindanao in the Philippines.

Pangasinan is written in the Latin script and is mainly spoken in the Pangasinan province of the Philippines.

Iloko is written in the Latin script and is mainly spoken in Northern Luzon, Philippines.

Waray is written in the Latin script and is mainly spoken in the Eastern Visayas region of the Philippines.

Haitian Creole is written in the Latin script and is one of the official languages of Haiti.

Papiamento is written in the Latin script and is mainly spoken in Caribbean regions such as Aruba and Curaçao.

Response format fixes: These models have fixed response format issues from previous versions, such as abnormal Markdown, mid-response truncation, and incorrect boxed output.

The open source Qwen3 models released in April 2025 do not support non-streaming output in `thinking mode`.

If an open source Qwen3 model is in `thinking mode` but does not output a chain-of-thought, it is billed at the `non-thinking mode` rate.

Thinking mode | Non-thinking mode | Usage

Global

In the Global deployment mode, endpoints and data storage are located in the US (Virginia) region, and inference compute resources are dynamically scheduled globally.

Model	Mode	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost	Free quota (Note)
Model	Mode	(Tokens)				(Million tokens)		Free quota (Note)
qwen3-next-80b-a3b-thinking	Thinking only	131,072	126,976	81,920	32,768	$0.15	$1.2	No free quota
qwen3-next-80b-a3b-instruct	Non-thinking only		129,024	-		$0.15	$1.2
qwen3-235b-a22b-thinking-2507	Thinking only		126,976	81,920		$0.23	$2.30
qwen3-235b-a22b-instruct-2507	Non-thinking only		129,024	-		$0.23	$0.92
qwen3-30b-a3b-thinking-2507	Thinking only		126,976	81,920		$0.20	$2.40
qwen3-30b-a3b-instruct-2507	Non-thinking only		129,024	-		$0.20	$0.80
qwen3-235b-a22b	Non-thinking		129,024	-	16,384	$0.70	$2.8
qwen3-235b-a22b	Thinking		98,304	38,912		$0.70	$8.40
qwen3-32b	Non-thinking		129,024	-		$0.16	$0.64
qwen3-32b	Thinking		98,304	38,912		$0.16	$0.64
qwen3-30b-a3b	Non-thinking		129,024	-		$0.20	$0.80
qwen3-30b-a3b	Thinking		98,304	38,912		$0.20	$2.40
qwen3-14b	Non-thinking		129,024	-	8,192	$0.35	$1.40
qwen3-14b	Thinking		98,304	38,912		$0.35	$4.2
qwen3-8b	Non-thinking		129,024	-		$0.18	$0.70
qwen3-8b	Thinking		98,304	38,912		$0.18	$2.1

International

In the International deployment mode, endpoints and data storage are located in the Singapore region, while inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model	Mode	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost	Free quota (Note)
Model	Mode	(Tokens)				(Million tokens)		Free quota (Note)
qwen3-next-80b-a3b-thinking	Thinking only	131,072	126,976	81,920	32,768	$0.15	$1.2	1 million tokens for each model Valid for 90 days after you activate Model Studio
qwen3-next-80b-a3b-instruct	Non-thinking only		129,024	-		$0.15	$1.2
qwen3-235b-a22b-thinking-2507	Thinking only		126,976	81,920		$0.23	$2.3
qwen3-235b-a22b-instruct-2507	Non-thinking only		129,024	-		$0.23	$0.92
qwen3-30b-a3b-thinking-2507	Thinking only		126,976	81,920		$0.2	$2.4
qwen3-30b-a3b-instruct-2507	Non-thinking only		129,024	-		$0.2	$0.8
qwen3-235b-a22b This model and the following models were released in April 2025.	Non-thinking		129,024	-	16,384	$0.7	$2.8
	Thinking		98,304	38,912		$0.7	$8.4
qwen3-32b	Non-thinking		129,024	-		$0.16	$0.64
qwen3-32b	Thinking		98,304	38,912		$0.16	$0.64
qwen3-30b-a3b	Non-thinking		129,024	-		$0.2	$0.8
qwen3-30b-a3b	Thinking		98,304	38,912		$0.2	$2.4
qwen3-14b	Non-thinking		129,024	-	8,192	$0.35	$1.4
qwen3-14b	Thinking		98,304	38,912		$0.35	$4.2
qwen3-8b	Non-thinking		129,024	-		$0.18	$0.7
qwen3-8b	Thinking		98,304	38,912		$0.18	$2.1
qwen3-4b	Non-thinking		129,024	-		$0.11	$0.42
qwen3-4b	Thinking		98,304	38,912			$1.26
qwen3-1.7b	Non-thinking	32,768	30,720	-			$0.42
qwen3-1.7b	Thinking		28,672	The total value cannot exceed 30,720.			$1.26
qwen3-0.6b	Non-thinking		30,720	-			$0.42
qwen3-0.6b	Thinking		28,672	The total value cannot exceed 30,720.			$1.26

Mainland China

In Mainland China deployment mode, the endpoint and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Mode	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost	Free quota (Note)
Model	Mode	(Tokens)				(Million tokens)		Free quota (Note)
qwen3-next-80b-a3b-thinking	Thinking only	131,072	126,976	81,920	32,768	$0.144	$1.434	No free quota
qwen3-next-80b-a3b-instruct	Non-thinking only		129,024	-		$0.144	$0.574
qwen3-235b-a22b-thinking-2507	Thinking only		126,976	81,920		$0.287	$2.868
qwen3-235b-a22b-instruct-2507	Non-thinking only		129,024	-		$0.287	$1.147
qwen3-30b-a3b-thinking-2507	Thinking only		126,976	81,920		$0.108	$1.076
qwen3-30b-a3b-instruct-2507	Non-thinking only		129,024	-		$0.108	$0.431
qwen3-235b-a22b	Non-thinking		129,024	-	16,384	$0.287	$1.147
qwen3-235b-a22b	Thinking		98,304	38,912		$0.287	$2.868
qwen3-32b	Non-thinking		129,024	-		$0.287	$1.147
qwen3-32b	Thinking		98,304	38,912		$0.287	$2.868
qwen3-30b-a3b	Non-thinking		129,024	-		$0.108	$0.431
qwen3-30b-a3b	Thinking		98,304	38,912		$0.108	$1.076
qwen3-14b	Non-thinking		129,024	-	8,192	$0.144	$0.574
qwen3-14b	Thinking		98,304	38,912		$0.144	$1.434
qwen3-8b	Non-thinking		129,024	-		$0.072	$0.287
qwen3-8b	Thinking		98,304	38,912		$0.072	$0.717
qwen3-4b	Non-thinking		129,024	-		$0.044	$0.173
qwen3-4b	Thinking		98,304	38,912			$0.431
qwen3-1.7b	Non-thinking	32,768	30,720	-			$0.173
qwen3-1.7b	Thinking		28,672	The total value cannot exceed 30,720.			$0.431
qwen3-0.6b	Non-thinking		30,720	-			$0.173
qwen3-0.6b	Thinking		28,672	The total value cannot exceed 30,720.			$0.431

QwQ - Open source

The QwQ reasoning model is trained on Qwen2.5-32B. Reinforcement learning has significantly improved its inference capabilities. Core metrics for math and code (AIME 24/25, LiveCodeBench) and some general metrics (IFEval, LiveBench) are comparable to the full-power version of DeepSeek-R1. All metrics significantly exceed those of DeepSeek-R1-Distill-Qwen-32B, which is also based on Qwen2.5-32B. Usage | API reference

Note

Model	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost
Model	(Tokens)				(Million tokens)
qwq-32b	131,072	98,304	32,768	8,192	$0.287	$0.861

QwQ-Preview

The qwq-32b-preview model is an experimental research model developed by the Qwen team in 2024. It focuses on enhancing AI reasoning capabilities, especially in math and programming. For more information about the limitations of the qwq-32b-preview model, see the QwQ official blog. Usage | API reference | Try it online

Note

Model	Context window	Max input	Max output	Input cost	Output cost
Model	(Tokens)			(Million tokens)
qwq-32b-preview	32,768	30,720	16,384	$0.287	$0.861

Qwen2.5

Qwen2.5 is a series of Qwen large language models that includes base and instruction-tuned language models with parameter sizes ranging from 7 billion to 72 billion. Qwen2.5 includes the following improvements over Qwen2:

It is pre-trained on our latest large-scale dataset, which contains up to 18 trillion tokens.
Pre-training with specialized expert models has significantly increased the model's knowledge and greatly improved its coding and math capabilities.
It shows significant improvements in following instructions, generating long text (over 8K tokens), understanding structured data (such as tables), and generating structured output (especially JSON). It is also more resilient to diverse system prompts, which enhances the implementation of role-playing and conditional settings for chatbots.
It supports over 29 languages, including Chinese, English, French, Spanish, Portuguese, German, Italian, Russian, Japanese, Korean, Vietnamese, Thai, and Arabic.

Usage | API reference | Try it online

International

Model	Context window	Max input	Max output	Input cost	Output cost	Free quota
	(Tokens)			(Million tokens)
qwen2.5-14b-instruct-1m	1,008,192	1,000,000	8,192	$0.805	$3.22	1 million tokens each Valid for 90 days after you activate Model Studio.
qwen2.5-7b-instruct-1m				$0.368	$1.47
qwen2.5-72b-instruct	131,072	129,024		$1.4	$5.6
qwen2.5-32b-instruct				$0.7	$2.8
qwen2.5-14b-instruct				$0.35	$1.4
qwen2.5-7b-instruct				$0.175	$0.7

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Context window	Max input	Max output	Input cost	Output cost
Model	(Tokens)			(Million tokens)
qwen2.5-14b-instruct-1m	1,000,000	1,000,000	8,192	$0.144	$0.431
qwen2.5-7b-instruct-1m	1,000,000	1,000,000		$0.072	$0.144
qwen2.5-72b-instruct	131,072	129,024		$0.574	$1.721
qwen2.5-32b-instruct				$0.287	$0.861
qwen2.5-14b-instruct				$0.144	$0.431
qwen2.5-7b-instruct				$0.072	$0.144
qwen2.5-3b-instruct	32,768	30,720		$0.044	$0.130
qwen2.5-1.5b-instruct				Free for a limited time
qwen2.5-0.5b-instruct				Free for a limited time

QVQ

The qvq-72b-preview model is an experimental research model developed by the Qwen team. It focuses on enhancing visual reasoning capabilities, especially in mathematical reasoning. For more information about the limitations of the qvq-72b-preview model, see the QVQ official blog.Usage | API reference

To have the model output its thinking process before the final answer, you can use the commercial version of the QVQ model.

Note

Model

Context window

Max input

Max output

Input cost

Output cost

(Tokens)

(Million tokens)

qvq-72b-preview

32,768

16,384

Max 16,384 tokens per image

16,384

$1.721

$5.161

Qwen-Omni

This is a new multimodal large model for understanding and generation, trained on Qwen2.5. It supports text, image, speech, and video inputs, and can generate text and speech simultaneously in a stream. Its multimodal content understanding speed is significantly improved.Usage | API reference

International

Model

Context window

Max input

Max output

Free quota

(Note)

(Tokens)

qwen2.5-omni-7b

32,768

30,720

2,048

1 million tokens (regardless of modality)

Valid for 90 days after activating Model Studio.

After the free quota is used up, the following billing rules apply to inputs and outputs:

Input	Price (Million tokens)
Text	$0.10
Audio	$6.76
Image/Video	$0.28

Output

Price (Million tokens)

Text

$0.40 (if the input contains only text)

$0.84 (if the input contains images, audio, or video)

Text+Audio

$13.51 (for the audio component)

The text portion of the output is not billed.

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Context window	Max input	Max output
Model	(Tokens)
qwen2.5-omni-7b	32,768	30,720	2,048

The billing rules for inputs and outputs are as follows:

Input	Price (Million tokens)
Text	$0.087
Audio	$5.448
Image/Video	$0.287

Output

Price (Million tokens)

Text

$0.345 (for text-only input)

$0.861 (if the input includes images, audio, or video)

Text+Audio

$10.895 (for the audio portion)

The text portion of the output is not billed.

Qwen3-Omni-Captioner

Qwen3-Omni-Captioner is an open-source model based on Qwen3-Omni. Without any prompts, it automatically generates accurate and comprehensive descriptions for complex audio, such as speech, ambient sounds, music, and sound effects. It can identify speaker emotions, musical elements (such as style and instruments), and sensitive information, making it suitable for applications such as audio content analysis, security audits, intent recognition, and audio editing. Usage | API reference

International

In the International deployment mode, the endpoint and data storage are located in the Singapore region, and inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model

Context window

Max input

Max output

Input cost

Output cost

Free quota

(Note)

(Tokens)

(Million tokens)

qwen3-omni-30b-a3b-captioner

65,536

32,768

$3.81

$3.06

1 million tokens

Valid for 90 days after you activate Alibaba Cloud Model Studio

Mainland China

In Mainland China deployment mode, the endpoint and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model

Context window

Max input

Max output

Input cost

Output cost

Free quota

(Note)

(Number of tokens)

(Million tokens)

qwen3-omni-30b-a3b-captioner

65,536

32,768

$2.265

$1.821

No free quota

Qwen-VL

This is the open-source version of Alibaba Cloud's Qwen-VL. Usage | API reference

The Qwen3-VL model offers significant improvements over Qwen2.5-VL:

Agent interaction: This model operates computer and mobile interfaces by recognizing Graphical User Interface (GUI) elements, understanding features, and calling tools to perform tasks. It achieves top-tier performance in evaluations such as OS World.
Visual encoding: This model generates code from images or videos. You can use it to create HTML, CSS, and JS code from design files or website screenshots.
Spatial intelligence: This model supports 2D and 3D positioning. It accurately determines object orientation, perspective changes, and occlusion relationships.
Long video understanding: This model understands video content up to 20 minutes long. It can pinpoint specific moments with second-level precision.
Deep thinking: This model excels at capturing details and analyzing cause and effect. It achieves top-tier performance in evaluations such as MathVista and MMMU.
OCR: This model supports 33 languages. It performs more stably in scenarios with complex lighting, blur, or tilt. The model significantly improves recognition accuracy for rare characters, ancient scripts, and technical terms.
Supported languages
The 33 supported languages are Chinese, Japanese, Korean, Indonesian, Vietnamese, Thai, English, French, German, Russian, Portuguese, Spanish, Italian, Swedish, Danish, Czech, Norwegian, Dutch, Finnish, Turkish, Polish, Swahili, Romanian, Serbian, Greek, Kazakh, Uzbek, Cebuano, Arabic, Urdu, Persian, Hindi / Devanagari, and Hebrew.

Global

In the global deployment mode, the endpoint and data storage are located in the US (Virginia) region, and inference compute resources are dynamically scheduled globally.

Model	Mode	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost Chain-of-thought and outputs
Model	Mode	(Tokens)				(Million tokens)
qwen3-vl-235b-a22b-thinking	Thinking only		126,976	81,920		$0.4	$4
qwen3-vl-235b-a22b-instruct	Non-Thinking only		129,024	-		$0.4	$1.6
qwen3-vl-32b-thinking	Thinking only	131,072	126,976	81,920	32,768	$0.16	$0.64
qwen3-vl-32b-instruct	Non-Thinking only		129,024	-		$0.16	$0.64
qwen3-vl-30b-a3b-thinking	Thinking only		126,976	81,920		$0.2	$2.4
qwen3-vl-30b-a3b-instruct	Non-Thinking only		129,024	-		$0.2	$0.8
qwen3-vl-8b-thinking	Thinking only		126,976	81,920		$0.18	$2.1
qwen3-vl-8b-instruct	Non-Thinking only		129,024	-		$0.18	$0.7

International

In the International deployment mode, endpoints and data storage are located in the Singapore region, and inference compute resources are dynamically scheduled globally (excluding Mainland China).

Model	Mode	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost chain-of-thought + Outputs	Free quota (Note)
Model	Mode	(Tokens)				(Million tokens)		Free quota (Note)
qwen3-vl-235b-a22b-thinking	Thinking mode		126,976	81,920		$0.4	$4	1 million tokens for each model Valid for 90 days after you activate Model Studio
qwen3-vl-235b-a22b-instruct	Non-thinking mode		129,024	-		$0.4	$1.6
qwen3-vl-32b-thinking	Thinking mode	131,072	126,976	81,920	32,768	$0.16	$0.64
qwen3-vl-32b-instruct	Non-thinking mode		129,024	-		$0.16	$0.64
qwen3-vl-30b-a3b-thinking	Thinking mode		126,976	81,920		$0.2	$2.4
qwen3-vl-30b-a3b-instruct	Non-thinking mode		129,024	-		$0.2	$0.8
qwen3-vl-8b-thinking	Thinking mode		126,976	81,920		$0.18	$2.1
qwen3-vl-8b-instruct	Non-thinking mode		129,024	-		$0.18	$0.7

More models

Model	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
	(Tokens)			(Million tokens)
qwen2.5-vl-72b-instruct	131,072	129,024 Max of 16,384 tokens per image	8,192	$2.8	$8.4	1 million tokens for each model Valid for 90 days after you activate Model Studio
qwen2.5-vl-32b-instruct				$1.4	$4.2
qwen2.5-vl-7b-instruct				$0.35	$1.05
qwen2.5-vl-3b-instruct				$0.21	$0.63

Mainland China

In Mainland China deployment mode, the endpoint and data storage are both located in the Beijing region, and inference compute resources are restricted to Mainland China.

Model	Mode	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost Chain-of-thought + Outputs	Free quota (Note)
Model	Mode	(Tokens)				(Million tokens)		Free quota (Note)
qwen3-vl-235b-a22b-thinking	Thinking only	131,072	126,976	81,920		$0.287	$2.867	No free quota
qwen3-vl-235b-a22b-instruct	Non-Thinking only	131,072	129,024	-		$0.287	$1.147
qwen3-vl-32b-thinking	Thinking only	131,072	126,976	81,920	32,768	$0.287	$2.868
qwen3-vl-32b-instruct	Non-Thinking only		129,024	-		$0.287	$1.147
qwen3-vl-30b-a3b-thinking	Thinking only		126,976	81,920		$0.108	$1.076
qwen3-vl-30b-a3b-instruct	Non-Thinking only		129,024	-		$0.108	$0.431
qwen3-vl-8b-thinking	Thinking only		126,976	81,920		$0.072	$0.717
qwen3-vl-8b-instruct	Non-Thinking only		129,024	-		$0.072	$0.287

More models

Model	Context window	Max input	Max output	Input cost	Output cost ($/million tokens)	Free quota (Note)
Model	Tokens			(Million tokens)		Free quota (Note)
qwen2.5-vl-72b-instruct	131,072	129,024 Max 16,384 per image	8,192	$2.294	$6.881	No free quota
qwen2.5-vl-32b-instruct				$1.147	$3.441
qwen2.5-vl-7b-instruct				$0.287	$0.717
qwen2.5-vl-3b-instruct				$0.173	$0.517
qwen2-vl-72b-instruct	32,768	30,720 Max 16,384 per image	2,048	$2.294	$6.881

Qwen-Math

This is a language model built on the Qwen model that is specialized for solving mathematical problems. Qwen2.5-Math supports Chinese and English and integrates multiple reasoning methods, such as Chain of Thought (CoT), Program of Thought (PoT), and Tool-Integrated Reasoning (TIR). Usage | API reference | Try it online

Note

Model	Context	Max input	Max output	Input cost	Output cost
	(Tokens)			(Million tokens)
qwen2.5-math-72b-instruct	4,096	3,072	3,072	$0.574	$1.721
qwen2.5-math-7b-instruct				$0.144	$0.287
qwen2.5-math-1.5b-instruct				Free for a limited time

Qwen-Coder

Qwen-Coder is an open-source code model from the Qwen series. The latest Qwen3-Coder series has powerful coding agent capabilities. It excels at tool calling, environment interaction, and autonomous programming. The model combines excellent coding skills with general-purpose capabilities. Usage | API reference

Global

In Global Deployment Mode, endpoints and data storage are both located in the US (Virginia) region, and inference compute resources are dynamically scheduled globally.

Model	Context window	Max input	Max output	Input cost	Output cost
	(Tokens)			(Million tokens)
qwen3-coder-480b-a35b-instruct	262,144	204,800	65,536	Tiered pricing applies. For more information, see the description below this table.
qwen3-coder-30b-a3b-instruct

qwen3-coder-480b-a35b-instruct and qwen3-coder-30b-a3b-instruct use tiered pricing based on the number of input tokens per request.

Model	Input tokens per request	Input cost (Million tokens)	Output cost (Million tokens)
qwen3-coder-480b-a35b-instruct	Up to 32K	$1.5	$7.5
	32 K < Token ≤ 128 K	$2.7	$13.5
	128 KB < Token ≤ 200 KB	$4.5	$22.5
qwen3-coder-30b-a3b-instruct	Up to 32K	$0.45	$2.25
	32 K < Token ≤ 128 K	$0.75	$3.75
	128 K < Token ≤ 200 K	$1.2	$6

International

Model	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
	(Tokens)
qwen3-coder-480b-a35b-instruct	262,144	204,800	65,536	Tiered pricing, see the description below the table.		1 million tokens each Validity: 90 days after you activate Model Studio
qwen3-coder-30b-a3b-instruct

qwen3-coder-480b-a35b-instruct and qwen3-coder-30b-a3b-instruct use tiered pricing based on the number of input tokens per request.

Model	Input tokens per request	Input cost (Million tokens)	Output cost (Million tokens)
qwen3-coder-480b-a35b-instruct	0 < tokens ≤ 32K	$1.5	$7.5
	32K < tokens ≤ 128K	$2.7	$13.5
	128K < tokens ≤ 200K	$4.5	$22.5
qwen3-coder-30b-a3b-instruct	0 < tokens ≤ 32K	$0.45	$2.25
	32K < tokens ≤ 128K	$0.75	$3.75
	128K < tokens ≤ 200K	$1.2	$6

Mainland China

In Mainland China deployment mode, endpoints and data storage are both located in the Beijing region, and inference compute resources are limited to Mainland China.

Model	Context window	Max input	Max output	Input cost	Output cost
	Tokens			(Million tokens)
qwen3-coder-480b-a35b-instruct	262,144	204,800	65,536	Tiered pricing applies. See the notes below the table.
qwen3-coder-30b-a3b-instruct
qwen2.5-coder-32b-instruct	131,072	129,024	8,192	$0.287	$0.861
qwen2.5-coder-14b-instruct
qwen2.5-coder-7b-instruct				$0.144	$0.287
qwen2.5-coder-3b-instruct	32,768	30,720		Limited-time free trial
qwen2.5-coder-1.5b-instruct
qwen2.5-coder-0.5b-instruct

qwen3-coder-480b-a35b-instruct and qwen3-coder-30b-a3b-instruct use tiered pricing based on the number of input tokens per request.

Model	Input tokens per request	Input cost (Million tokens)	Output cost (Million tokens)
qwen3-coder-480b-a35b-instruct	Up to 32K	$0.861	$3.441
	32K to 128K	$1.291	$5.161
	128K to 200K	$2.151	$8.602
qwen3-coder-30b-a3b-instruct	Up to 32K	$0.216	$0.861
	32K to 128K	$0.323	$1.291
	128K to 200K	$0.538	$2.151

Text generation - Third-party

DeepSeek

DeepSeek is a large language model from DeepSeek AI. API reference | Try it online

Note

Model	Context window	Max input	Max chain-of-thought	Max response	Input cost	Output cost
	(Tokens)				(Million tokens)
deepseek-v3.2 685B full-power version	131,072	98,304	32,768	65,536	$0.287	$0.431
deepseek-v3.2-exp 685B full-power version
deepseek-v3.1 685B full-power version					$0.574	$1.721
deepseek-r1 685B full-power version				16,384		$2.294
deepseek-r1-0528 685B full-power version
deepseek-v3 671B full-power version		131,072	N/A		$0.287	$1.147
deepseek-r1-distill-qwen-1.5b Based on Qwen2.5-Math-1.5B	32,768	32,768	16,384	16,384	Free trial for a limited time
deepseek-r1-distill-qwen-7b Based on Qwen2.5-Math-7B					$0.072	$0.144
deepseek-r1-distill-qwen-14b Based on Qwen2.5-14B					$0.144	$0.431
deepseek-r1-distill-qwen-32b Based on Qwen2.5-32B					$0.287	$0.861
deepseek-r1-distill-llama-8b Based on Llama-3.1-8B					Free trial for a limited time
deepseek-r1-distill-llama-70b Based on Llama-3.3-70B

Kimi

Kimi-K2 is a large language model launched by Moonshot AI. It has excellent coding and tool-calling capabilities. Usage | Try it online

Note

Model	Context window	Max input	Max chain-of-thought	Max response	Input price	Output price
Model	(Tokens)				(Million tokens)
kimi-k2-thinking	262,144	229,376	32,768	16,384	$0.574	$2.294
Moonshot-Kimi-K2-Instruct	131,072	131,072	-	8,192	$0.574	$2.294

Image generation

Qwen-Image

The Qwen text-to-image model excels at rendering complex text, especially in Chinese and English. API reference

International

Model	Unit price	Free quota
Model	Unit price	Free quota
qwen-image-max Currently has the same capabilities as qwen-image-max-2025-12-30	$0.075/image	Free quota: 100 images for each model Validity period: Within 90 days after you activate Model Studio
qwen-image-max-2025-12-30	$0.075/image
qwen-image-plus Currently has the same capabilities as qwen-image	$0.03/image
qwen-image-plus-2026-01-09	$0.03/image
qwen-image	$0.035/image

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Unit price	Free quota
Model	Unit price	Free quota
qwen-image-max Currently has the same capabilities as qwen-image-max-2025-12-30	$0.071677/image	No free quota
qwen-image-max-2025-12-30	$0.071677/image
qwen-image-plus Currently has the same capabilities as qwen-image	$0.028671/image
qwen-image-plus-2026-01-09	$0.028671/image
qwen-image	$0.035/image

Input prompt

Output image

Healing-style hand-drawn poster featuring three puppies playing with a ball on lush green grass, adorned with decorative elements such as birds and stars. The main title “Come Play Ball!” is prominently displayed at the top in bold, blue cartoon font. Below it, the subtitle “Come [Show Off Your Skills]!” appears in green font. A speech bubble adds playful charm with the text: “Hehe, watch me amaze my little friends next!” At the bottom, supplementary text reads: “We get to play ball with our friends again!” The color palette centers on fresh greens and blues, accented with bright pink and yellow tones to highlight a cheerful, childlike atmosphere.

Qwen-Image-Edit

The Qwen image editing model supports precise text editing in Chinese and English. It also supports operations such as color adjustment, detail enhancement, style transfer, adding or removing objects, and changing positions and actions. These features enable complex editing of images and text. API reference

International

Model	Unit price	Free quota

qwen-image-edit-plus Currently has the same capabilities as qwen-image-edit-plus-2025-10-30	$0.03/image	Free quota: 100 images for each model Validity period: Within 90 days after you activate Model Studio
qwen-image-edit-plus-2025-12-15	$0.03/image
qwen-image-edit-plus-2025-10-30	$0.03/image
qwen-image-edit	$0.045/image

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Unit price	Free quota

qwen-image-edit-plus Currently has the same capabilities as qwen-image-edit-plus-2025-10-30	$0.028671/image	No free quota
qwen-image-edit-plus-2025-12-15	$0.028671/image
qwen-image-edit-plus-2025-10-30	$0.028671/image
qwen-image-edit	$0.043/image

dog_and_girl (1)

Original image

狗修改图

Make the person bend over and hold the dog's front paw.

Original image

Change the text on the letter blocks from 'HEALTH INSURANCE' to 'Tomorrow will be better'.

Original image

5out

Change the dotted shirt to a light blue shirt.

Original image

6out

Change the background to Antarctica.

Original image

7out

Create a cartoon-style profile picture of the person.

Original image

Remove the hair from the dinner plate.

Qwen-MT-Image

The Qwen image translation model supports translating text from images in 11 languages into Chinese or English. It accurately preserves the original layout and content information and provides custom features such as term definition, sensitive word filtering, and image entity detection. API reference

Note

Model	Unit price	Free quota

qwen-mt-image	$0.000431/image	No free quota

Original image

Japanese

Portuguese

Arabic

Tongyi - text-to-Image - Z-Image

Tongyi - text-to-image - Z-Image is a lightweight model that quickly generates high-quality images. The model supports Chinese and English text rendering, complex semantic understanding, various styles, and multiple resolutions and aspect ratios. API reference

International

Model

Unit price

Free quota (Note)

_{Validity period: Within 90 days after you activate Model Studio}

z-image-turbo

Prompt extension disabled (prompt_extend=false): $0.015/image

Prompt extension enabled (prompt_extend=true): $0.03/image

100 images

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model

Unit price

Free quota

z-image-turbo

Prompt extension disabled (prompt_extend=false): $0.01434/image

Prompt extension enabled (prompt_extend=true): $0.02868/image

No free quota

Input prompt

Output image

Photo of a stylish young woman with short black hair standing confidently in front of a vibrant cartoon-style mural wall. She wears an all-black outfit: a puffed bomber jacket with a ruffled collar, cargo shorts, fishnet tights, and chunky black Doc Martens, with a gold chain dangling from her waist. The background features four colorful comic-style panels: one reads “GRAND STAGE” and includes sneakers and a Gatorade bottle; another displays green Nike sneakers and a slice of pizza; the third reads “HARAJUKU st” with floating shoes; and the fourth shows a blue mouse riding a skateboard with the text “Takeshita WELCOME.” Dominant bright colors include yellow, teal, orange, pink, and green. Speech bubbles, halftone patterns, and playful characters enhance the urban street-art aesthetic. Daylight evenly illuminates the scene, and the ground beneath her feet is white tiled pavement. Full-body portrait, centered composition, slightly tilted stance, direct eye contact with the camera. High detail, sharp focus, dynamic framing.

b16c8008-83c1-4c80-ae22-786a2299bec3-1-转换自-png

Wan text-to-image

The Wan text-to-image model generates high-quality images from text. API reference | Try it online

Global

In Global deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are dynamically scheduled globally.

Model

Description

Unit price

Free quota (Note)

_{Validity period: Within 90 days after you activate Model Studio}

wan2.6-t2i Recommended

Wan 2.6. Supports new synchronous interfaces and lets you freely select dimensions within the constraints of total pixel area and aspect ratio.

$0.03/image

No free quota

International

Model	Description	Unit price	Free quota (Note) _{Validity period: Within 90 days after you activate Model Studio}
wan2.6-t2i `Recommended`	Wan 2.6. Supports new synchronous interfaces and lets you freely select dimensions within the constraints of total pixel area and aspect ratio.	$0.03/image	50 images
wan2.5-t2i-preview `Recommended`	Wan 2.5 preview. Removes single-side length limits and lets you freely select dimensions within the constraints of total pixel area and aspect ratio.	$0.03/image	50 images
wan2.2-t2i-plus	Wan 2.2 Professional Edition. Fully upgraded in creativity, stability, and realistic texture.	$0.05/image	100 images
wan2.2-t2i-flash	Wan 2.2 Flash Edition. Fully upgraded in creativity, stability, and realistic texture.	$0.025/image	100 images
wan2.1-t2i-plus	Wan 2.1 Professional Edition. Supports multiple styles and generates images with rich details.	$0.05/image	200 images
wan2.1-t2i-turbo	Wan 2.1 Turbo Edition. Supports multiple styles and offers fast generation speed.	$0.025/image	200 images

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Description	Unit price	Free quota (Note) _{Validity period: Within 90 days after you activate Model Studio}
wan2.6-t2i `Recommended`	Wan 2.6. Supports new synchronous interfaces and lets you freely select dimensions within the constraints of total pixel area and aspect ratio.	$0.028671/image	No free quota
wan2.5-t2i-preview `Recommended`	Wan 2.5 preview. Removes single-side length limits and lets you freely select dimensions within the constraints of total pixel area and aspect ratio.	$0.028671/image	No free quota
wan2.2-t2i-plus	Wan 2.2 Professional Edition. Fully upgraded in creativity, stability, and realistic texture.	$0.02007/image	No free quota
wan2.2-t2i-flash	Wan 2.2 Flash Edition. Fully upgraded in creativity, stability, and realistic texture.	$0.028671/image	No free quota
wanx2.1-t2i-plus	Wan 2.1 Professional Edition. Supports multiple styles and generates images with rich details.	$0.028671/image	No free quota
wanx2.1-t2i-turbo	Wan 2.1 Turbo Edition. Supports multiple styles and offers fast generation speed.	$0.020070/image	No free quota
wanx2.0-t2i-turbo	Wan 2.0 Turbo Edition. Excels at textured portraits and creative designs. It is cost-effective.	$0.005735/image	No free quota

Input prompt	Output image
A needle-felted Santa Claus holding a gift and a white cat standing next to him against a background of colorful gifts and green plants, creating a cute, warm, and cozy scene.

Wan2.6 image generation and editing

The Wan2.6 image generation model supports image editing and can generate outputs that contain both text and images to meet various generation and integration requirements. API reference.

Global

In Global deployment mode, the access point and data storage are in the US (Virginia) region, and inference computing resources are dynamically scheduled globally.

Model	Unit price	Free quota
wan2.6-image	$0.03/image	No free quota

International

In International deployment mode, the access point and data storage are in the Singapore region, and inference computing resources are dynamically scheduled globally (excluding Mainland China).

Model

Unit price

Free quota (Note)

_{Validity period: Within 90 days after you activate Model Studio}

wan2.6-image

$0.03/image

50 images

Mainland China

In Mainland China deployment mode, the access point and data storage are in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Unit price	Free quota
wan2.6-image	$0.028671/image	No free quota

Wan general image editing 2.5

The Wan2.5 general image editing model supports entity-consistent image editing and multi-image fusion. It accepts text, a single image, or multiple images as input. API reference.

International

Model

Unit price

Free quota (Note)

_{Validity period: Within 90 days after you activate Model Studio}

wan2.5-i2i-preview

$0.03/image

50 units

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Unit price	Free quota
wan2.5-i2i-preview	$0.028671/image	No free quota

Feature

Input example

Output image

Single-image editing

damotest2023_Portrait_photography_outdoors_fashionable_beauty_409ae3c1-19e8-4515-8e50-b3c9072e1282_2-转换自-png

a26b226d-f044-4e95-a41c-d1c0d301c30b-转换自-png

Change the floral dress to a vintage-style lace long dress with exquisite embroidery details on the collar and cuffs.

Multi-image fusion

p1028883

Place the alarm clock from Image 1 next to the vase on the dining table in Image 2.

Wan general image editing 2.1

The Wan2.1 general image editing model performs diverse image editing with simple instructions. It is suitable for scenarios such as outpainting, watermark removal, style transfer, image restoration, and image enhancement. Usage | API reference

Note

Model	Unit Price	Free Quota
wanx2.1-imageedit	$0.020070 per image	No free quota

The general image editing model currently supports the following features:

Model Features	Input image	Input prompt	Output image
Global stylization		French picture book style.
Local stylization		Change the house to a wooden plank style.
Instruction-based editing		Change the girl's hair to red.
Inpainting	Input image Masked image (The white area is the mask)	A ceramic rabbit holding a ceramic flower.	Output image
Text watermark removal		Remove the text from the image.
Outpainting		A green fairy.
Image super-resolution	Blurry image	Image super-resolution.	Clear image
Image colorization		Blue background, yellow leaves.
Line art to image		A living room in a minimalist Nordic style.
Placeholder Image		A cartoon character cautiously peeks out, spying on a brilliant blue gem inside the room.

OutfitAnyone

Compared to the basic version, the OutfitAnyone-Plus model offers improvements in image definition, clothing texture details, and logo restoration. However, it takes longer to generate images and is suitable for scenarios that are not time-sensitive. API reference | Try it online
OutfitAnyone-Image Parsing supports parsing model and clothing images, which can be used for pre-processing and post-processing of OutfitAnyone images. API reference

Note

Model	Description	Sample input	Sample output
aitryon-plus	OutfitAnyone-Plus
aitryon-parsing-v1	OutfitAnyone image parsing

OutfitAnyone pricing

Service	Model	Unit price	Discount	Tier
OutfitAnyone - Plus	aitryon-plus	$0.071677/image	None	None
OutfitAnyone - Image parsing	aitryon-parsing-v1	$0.000574/image	None	None

Video generation - Wan

Text-to-video

The Wan text-to-video model generates videos from a single sentence. The videos feature rich artistic styles and cinematic quality. API reference | Try it online

Global

In Global deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are dynamically scheduled globally.

Model

Description

Unit price

Free quota

wan2.6-t2v Recommended

Wan 2.6. Introduces a multi-shot narrative feature and supports automatic voiceover and the import of custom audio files.

720P: $0.1/second

1080P: $0.15/second

No free quota

International

Model	Description	Unit price	Free quota (Claim) _{Valid for 90 days after you activate Alibaba Cloud Model Studio}
wan2.6-t2v `Recommended`	Wan 2.6. Introduces a multi-shot narrative feature and supports automatic voiceover and the import of custom audio files.	720P: $0.10/second 1080P: $0.15/second	50 seconds
wan2.5-t2v-preview `Recommended`	Wan 2.5 preview. Supports automatic voiceover and custom audio file input.	480p: $0.05/second 720p: $0.10/second 1080p: $0.15/second	50 seconds
wan2.2-t2v-plus	Wan 2.2 Professional Edition. Significantly improved image detail and motion stability.	480p: $0.02/second 1080p: $0.10/second	50 seconds
wan2.1-t2v-turbo	Wan 2.1 Turbo Edition. Fast generation speed and balanced performance.	$0.036/second	200 seconds
wan2.1-t2v-plus	Wan 2.1 Professional Edition. Generates rich details and higher-quality images.	$0.10/second	200 seconds

US

In US deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are limited to the US.

Model

Description

Unit price

Free quota

wan2.6-t2v-us Recommended

Wan 2.6. Introduces a multi-shot narrative feature and supports automatic voiceover and the import of custom audio files.

720P: $0.1/second

1080P: $0.15/second

No free quota

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Description	Unit price	Free quota
wan2.6-t2v `REcommended`	Wan 2.6. Introduces a multi-shot narrative feature and supports automatic voiceover and the import of custom audio files.	720P: $0.086012/second 1080P: 0.143353/second	No free quota
wan2.5-t2v-preview `Recommended`	Wan 2.5 preview. Supports automatic voiceover and custom audio file input.	480p: $0.043006/second 720p: $0.086012/second 1080p: $0.143353/second	No free quota
wan2.2-t2v-plus	Wan 2.2 Professional Edition. Significantly improved image detail and motion stability.	480p: $0.02007/second 1080p: $0.100347/second	No free quota
wanx2.1-t2v-turbo	Faster generation speed and balanced performance.	$0.034405/second	No free quota
wanx2.1-t2v-plus	Generates richer details and higher-quality images.	$0.100347/second	No free quota

Input prompt

Output video (wan2.6, multi-shot video)

Shot from a low angle, in a medium close-up, with warm tones, mixed lighting (the practical light from the desk lamp blends with the overcast light from the window), side lighting, and a central composition. In a classic detective office, wooden bookshelves are filled with old case files and ashtrays. A green desk lamp illuminates a case file spread out in the center of the desk. A fox, wearing a dark brown trench coat and a light gray fedora, sits in a leather chair, its fur crimson, its tail resting lightly on the edge, its fingers slowly turning yellowed pages. Outside, a steady drizzle falls beneath a blue sky, streaking the glass with meandering streaks. It slowly raises its head, its ears twitching slightly, its amber eyes gazing directly at the camera, its mouth clearly moving as it speaks in a smooth, cynical voice: 'The case was cold, colder than a fish in winter. But every chicken has its secrets, and I, for one, intended to find them '.

Image-to-video - first frame

The Wan image-to-video model uses an input image as the first frame of a video. It then generates the rest of the video based on a prompt. The videos feature rich artistic styles and cinematic quality. API reference | Try it online

Global

In Global deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are dynamically scheduled globally.

Model

Description

Unit price

Free quota

wan2.6-i2v Recommended

Wan 2.6. Introduces a multi-shot narrative feature and supports automatic voiceover and the import of custom audio files.

720P: $0.1/second

1080P: $0.15/second

No free quota

International

Model	Description	Unit price	Free quota (Note) _{Validity: Within 90 days after you activate Alibaba Cloud Model Studio}
wan2.6-i2v `Recommended`	Wan 2.6. Introduces a multi-shot narrative feature and supports automatic voiceover and the import of custom audio files.	720P: $0.10/seconds 1080P: $0.15/seconds	50seconds
wan2.5-i2v-preview `Recommended`	Wan 2.5 preview. Supports automatic dubbing and custom audio file uploads.	480P: $0.05/second 720P: $0.10/second 1080P: $0.15/second	50 seconds
wan2.2-i2v-flash	Wan 2.2 Flash Edition. Delivers extremely fast generation speed with significant improvements in visual detail and motion stability.	480P: $0.015/second 720P: $0.036/second	50 seconds
wan2.2-i2v-plus	Wan 2.2 Professional Edition. Delivers significant improvements in visual detail and motion stability.	480P: $0.02/second 1080P: $0.10/second	50 seconds
wan2.1-i2v-turbo	Wan 2.1 Turbo Edition. Fast generation speed with balanced performance.	$0.036/second	200 seconds
wan2.1-i2v-plus	Wan 2.1 Professional Edition. Generates rich details and produces higher-quality, more textured visuals.	$0.10/second	200 seconds

US

In US deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are limited to the US.

Model

Description

Unit price

Free quota

wan2.6-i2v-us Recommended

Wan 2.6. Introduces a multi-shot narrative feature and supports automatic voiceover and the import of custom audio files.

720P: $0.1/second

1080P: $0.15/second

No free quota

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Description	Unit price	Free quota
wan2.6-i2v `Recommended`	Wan 2.6. Introduces a multi-shot narrative feature and supports automatic voiceover and the import of custom audio files.	720P: $0.086012/second 1080P: $0.143353/second	No free quota
wan2.5-i2v-preview `Recommended`	Wan 2.5 preview. Supports automatic dubbing and custom audio file uploads.	480P: $0.043006/second 720P: $0.086012/second 1080P: $0.143353/second	No free quota
wan2.2-i2v-plus	Wan 2.2 Professional Edition. Delivers significant improvements in visual detail and motion stability.	480P: $0.02007/second 1080P: $0.100347/second	No free quota
wanx2.1-i2v-turbo	Wan 2.1 Turbo Edition. Fast generation speed with balanced performance.	$0.034405/second	No free quota
wanx2.1-i2v-plus	Wan 2.1 Professional Edition. Generates rich details and produces higher-quality, more textured visuals.	$0.100347/second	No free quota

Input first frame image and audio

Output video (wan2.6, multi-shot video)

rap-转换自-png

Input audio:

Input prompt: A scene of urban fantasy art. A dynamic graffiti art character. A boy painted with spray paint comes to life from a concrete wall. He sings an English rap song at a very fast pace while striking a classic, energetic rapper pose. The scene is set under an urban railway bridge at night. The lighting comes from a single streetlight, creating a cinematic atmosphere full of high energy and amazing detail. The audio of the video consists entirely of his rap, with no other dialogue or noise.

Image-to-video - first and last frames

The Wan first-and-last-frame video model generates a smooth, dynamic video from a prompt. You only need to provide the first and last frame images. The videos feature rich artistic styles and cinematic quality. API reference | Try it online

International

Model

Unit price

Free quota (Note)

^{Validity: 90 days after you activate Model Studio}

wan2.2-kf2v-flash

480P: $0.015/second

720P: $0.036/second

1080P: $0.07/second

50 seconds

wan2.1-kf2v-plus

$0.10/second

200 seconds

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model

Unit price

Free quota (Note)

wan2.2-kf2v-flash

480P: $0.014335/second

720P: $0.028671/second

1080P: $0.068809/second

No free quota

wanx2.1-kf2v-plus

$0.100347/second

No free quota

Example input			Output video
First frame	Last frame	Prompt	Output video
		In a realistic style, the camera starts at eye level on a small black cat looking up at the sky, then gradually moves upward to a top-down shot that focuses on the cat's curious eyes.

Reference-to-video

The Wan reference-to-video model uses a character's appearance and voice from an input video and a prompt to generate a new video that maintains character consistency. API reference

Billing rule: Both input and output videos are billed by the second. Failed jobs are not billed and do not consume the free quota.

The billable duration of the input video does not exceed 5 seconds. For more information, see Billing and rate limits.
The billable duration of the output video is the duration in seconds of the successfully generated video.

Global

In Global deployment mode, the access point and data storage are located in the US (Virginia) region, and inference computing resources are dynamically scheduled globally.

Model

Input price

Output price

Free quota (Note)

wan2.6-r2v

720P: $0.086012/second

1080P: $0.143353/second

720P: $0.1/second

1080P: $0.15/second

No free quota

International

Model

Input price

Output price

Free quota (Note)

wan2.6-r2v

720P: $0.10/second

1080P: $0.15/second

720P: $0.10/second

1080P: $0.15/second

50 seconds

Validity: 90 days after you activate Model Studio

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model

Input price

Output price

Free quota (Note)

wan2.6-r2v

720P: $0.086012/second

1080P: $0.143353/second

720P: $0.086012/second

1080P: $0.143353/second

No free quota

General video editing

The Wan general video editing model supports multimodal inputs, including text, images, and videos. It can perform video generation and general editing tasks. API reference | Try it online

International

Model

Unit price

Free quota (Note)

wan2.1-vace-plus

$0.1/s

50 seconds

Validity: Valid for 90 days after Model Studio activation.

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Unit price	Free quota (Note)
wanx2.1-vace-plus	$0.100347/s	No free quota

The general video editing model supports the following features:

Feature	Input reference image	Input prompt	Output video
Multi-image reference	Reference image 1 (reference entity) Reference image 2 (reference background)	In the video, a girl gracefully walks out from a misty, ancient forest. Her steps are light, and the camera captures her every nimble moment. When the girl stops and looks around at the lush woods, a smile of surprise and joy blossoms on her face. This scene, frozen in a moment of interplay between light and shadow, records her wonderful encounter with nature.	Output video
Video repainting		The video shows a black steampunk-style car driven by a gentleman. The car is decorated with gears and copper pipes. The background features a steam-powered candy factory and retro elements, creating a vintage and playful scene.
Local editing	Input video Input mask image (The white area indicates the editing area)	The video shows a Parisian-style French cafe where a lion in a suit is elegantly sipping coffee. It holds a coffee cup in one hand, taking a gentle sip with a relaxed expression. The cafe is tastefully decorated, with soft hues and warm lighting illuminating the area where the lion is.	The content in the editing area is modified based on the prompt.
Video extension	Input first clip (1 second)	A dog wearing sunglasses is skateboarding on the street, 3D cartoon.	Output extended video (5 seconds)
Video outpainting		An elegant lady is passionately playing the violin, with a full symphony orchestra behind her.

Wan - digital human

This feature generates natural-looking videos of people speaking, singing, or performing, based on a single character image and an audio file. To use this feature, you can call the following models in sequence. wan2.2-s2v image detection | wan2.2-s2v video generation

Note

Model	Description	Unit price

wan2.2-s2v-detect	Checks if an input image meets requirements, such as sufficient definition, a single person, and a frontal view.	$0.000574/image
wan2.2-s2v	Generates a dynamic video of a person from a valid image and an audio clip.	480p: $0.071677/second 720p: $0.129018/second

Sample input

Output video

p1001125-转换自-jpeg

Input audio:

Wan - animate image

Available in standard and professional modes. The model transfers the actions and expressions from a reference video to a character image, generating a video that animates the character from the image. API reference.

International

Model	Service	Description	Unit price	Free quota (View)
wan2.2-animate-move	Standard mode `wan-std`	A cost-effective service with fast generation speeds. Suitable for basic needs, such as simple animation demos.	$0.12/second	The total time for both patterns is 50 seconds.
wan2.2-animate-move	Professional mode `wan-pro`	Delivers high animation smoothness and natural transitions for actions and expressions. The output resembles a live-action video.	$0.18/second	The total time for both patterns is 50 seconds.

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Service	Description	Unit price	Free quota (View)
wan2.2-animate-move	Standard mode `wan-std`	Fast generation. Ideal for basic needs, such as simple animation demos. Cost-effective.	$0.06/second	No free quota
wan2.2-animate-move	Professional mode `wan-pro`	Provides high-quality, smooth animation with natural transitions for actions and expressions. The output is similar to a live-action video.	$0.09/second	No free quota

Character image	Reference video	Standard video	Output Video (Professional Mode)

Wan - video character swap

Available in standard and professional modes. The model replaces the main character in a video with a character from an image. It preserves the original video's scene, lighting, and hue. API reference.

International

Model	Service	Description	Unit price	Free quota (View)
wan2.2-animate-mix	Standard mode `wan-std`	Generates animations quickly. Ideal for basic requirements, such as simple demos. Highly cost-effective.	$0.18/s	The combined duration of both services is 50 seconds.
wan2.2-animate-mix	Professional mode `wan-pro`	Produces highly smooth animations with natural transitions for actions and expressions. The result closely resembles a live-action video.	$0.26/s	The combined duration of both services is 50 seconds.

Mainland China

In Mainland China deployment mode, the access point and data storage are located in the Beijing region, and inference computing resources are limited to Mainland China.

Model	Service	Description	Unit price	Free quota (View)
wan2.2-animate-mix	Standard mode `wan-std`	Generates animations quickly. Ideal for basic requirements, such as simple demos. Highly cost-effective.	$0.09/s	No free quota
wan2.2-animate-mix	Professional mode `wan-pro`	Produces highly smooth animations with natural transitions for actions and expressions. The result closely resembles a live-action video.	$0.13/s	No free quota

Character image	Reference video	Standard output video	Professional output video

AnimateAnyone

This feature generates character motion videos based on a character image and a motion template. To use this feature, you can call the following three models in sequence. AnimateAnyone image detection API details | AnimateAnyone motion template generation | AnimateAnyone video generation API details

Note

Model	Description	Unit price
animate-anyone-detect-gen2	Detects whether an input image meets the requirements.	$0.000574/image
animate-anyone-template-gen2	Extracts character motion from a video and generates a motion template.	$0.011469/second
animate-anyone-gen2	Generates a character action video from a character image and an action template.	$0.011469/second

Input: Character image	Input: Motion video	Outputs (generated from the image background)	Outputs Generated by Video Background

Note

The preceding example was generated by the Tongyi App, which integrates AnimateAnyone.
The content generated by the AnimateAnyone model is video only and does not include audio.

EMO

This feature generates dynamic portrait videos based on a portrait image and a human voice audio file. To use this feature, you can call the following models in sequence. EMO image detection | EMO video generation

Note

Model	Description	Unit price
emo-detect-v1	Detects whether an input image meets the required specifications. This model can be called directly without deployment.	$0.000574/image
emo-v1	Generates a dynamic portrait video. This model can be called directly without deployment.	1:1 aspect ratio video: $0.011469/second 3:4 aspect ratio video: $0.022937/second

Input: Portrait image and human voice audio file

Output: Dynamic portrait video

Portrait:

上春山

Human voice audio: See the video on the right.

Character video:

Style level: active ("style_level": "active")

LivePortrait

This model quickly and efficiently generates dynamic portrait videos based on a portrait image and a human voice audio file. Compared to the EMO model, it generates videos faster and at a lower cost, but the quality is not as good. To use this feature, you can call the following two models in sequence. LivePortrait image detection | LivePortrait video generation

Note

Model	Description	Unit price
liveportrait-detect	Detects whether an input image meets the requirements.	$0.000574/image
liveportrait	Generates a dynamic portrait video.	$0.002868/second

Input: Portrait image and voice audio

Output: Animated portrait video

Portrait image:

Emoji男孩

Voice audio: Sourced from the video on the right.

Portrait video:

Emoji

This feature generates dynamic face videos based on a face image and preset facial motion templates. This capability can be used for scenarios such as creating emojis and generating video materials. To use this feature, you can call the following models in sequence. Emoji image detection | Emoji video generation

Note

Model	Description	Unit price
emoji-detect-v1	Detects whether an input image meets specified requirements.	$0.000574/image
emoji-v1	Generates custom emojis based on a portrait image and a specified emoji template.	$0.011469/second

Input: Portrait image	Output: Dynamic portrait video
	Parameter for the "Happy" emoji template: ("input.driven_id": "mengwa_kaixin")

VideoRetalk

This feature generates a video where the character's lip movements match the input audio, based on a character video and a human voice audio file. To use this feature, you can call the following model. API reference

Note

Model	Description	Unit price
videoretalk	Synchronizes a character's lip movements with input audio to generate a new video.	$0.011469/second

Video style transform

This model generates videos in different styles that match the semantic description of user-input text, or restyles a user-input video. API reference

Note

Model	Description	Unit price

video-style-transform	Transforms an input video into styles such as Japanese comic and American comic.	720P	$0.071677/second
		540P	$0.028671/second

Input video	Output video (Manga style)

Speech synthesis (text-to-speech)

Qwen speech synthesis

This feature supports multilingual mixed-text input and provides streaming audio output. Usage | API reference

International

In the international deployment mode, the endpoint and data storage are located in the Singapore region. Inference computing resources are dynamically scheduled globally, excluding Mainland China.

Model	Version	Price	Maximum input characters	Supported languages	Free quota (Note)
qwen3-tts-flash Same capabilities as qwen3-tts-flash-2025-09-18.	Stable	$0.10 per 10,000 characters	600	Chinese (Mandarin, Beijing, Shanghai, Sichuan, Nanjing, Shaanxi, Minnan, Tianjin, Cantonese), English, Spanish, Russian, Italian, French, Korean, Japanese, German, Portuguese	2,000 characters if you activate Model Studio before 00:00 on November 13, 2025. 10,000 characters if you activate Model Studio on or after 00:00 on November 13, 2025. Valid for 90 days after you activate Model Studio.
qwen3-tts-flash-2025-11-27	Snapshot				10,000 characters Valid for 90 days after you activate Model Studio.
qwen3-tts-flash-2025-09-18	Snapshot				2,000 characters if you activate Model Studio before 00:00 on November 13, 2025. 10,000 characters if you activate Model Studio on or after 00:00 on November 13, 2025. Valid for 90 days after you activate Model Studio.

Billing is based on the number of input characters. The calculation rules are as follows:

Each Chinese character, including simplified and traditional Chinese, Japanese Kanji, and Korean Hanja, counts as 2 characters.
Any other character, such as an English letter, a punctuation mark, or a space, counts as 1 character.

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the China (Beijing) region, and inference computing resources are restricted to Mainland China.

Qwen3-TTS-Flash

Model	Version	Price	Max input characters	Supported languages	Free quota (Note)
qwen3-tts-flash Same capabilities as qwen3-tts-flash-2025-09-18.	Stable	$0.114682/10,000 characters	600	Chinese (Mandarin, Beijing, Shanghai, Sichuan, Nanjing, Shaanxi, Minnan, Tianjin, Cantonese), English, Spanish, Russian, Italian, French, Korean, Japanese, German, Portuguese	No free quota
qwen3-tts-flash-2025-11-27	Snapshot
qwen3-tts-flash-2025-09-18	Snapshot

Billing is based on the number of input characters. The calculation rules are as follows:

Each Chinese character, including simplified and traditional Chinese, Japanese Kanji, and Korean Hanja, counts as 2 characters.
Any other character, such as an English letter, a punctuation mark, or a space, counts as 1 character.

Qwen-TTS

Model	Version	Context window	Max input	Max output	Input cost	Output cost	Free quota (Note)
		(Tokens)			(per 1,000 tokens)
qwen-tts This model has the same capabilities as qwen-tts-2025-04-10.	Stable	8,192	512	7,680	$0.230	$1.434	No free quota
qwen-tts-latest This model always has the same capabilities as the latest snapshot version.	Latest
qwen-tts-2025-05-22	Snapshot
qwen-tts-2025-04-10

Audio is converted to tokens at a rate of 50 tokens per second. Audio clips shorter than 1 second are billed as 50 tokens.

Qwen real-time speech synthesis

This feature supports streaming text input and streaming audio output. It automatically adjusts the speech rate based on the text content and punctuation. Usage | API reference

Qwen3-TTS-VD-Realtime supports real-time speech synthesis with voice design voices but does not support default voices.

Qwen3-TTS-VC-Realtime supports real-time speech synthesis with cloned voices but does not support default voices.

Qwen3-TTS-Flash-Realtime and Qwen-TTS-Realtime support only default voices and do not support cloned or designed voices.

International

Qwen3-TTS-VD-Realtime

Model

Version

Price

Supported languages

Free quota (Note)

qwen3-tts-vd-realtime-2025-12-16

Snapshot

$0.143353/10,000 characters

Chinese, English, Spanish, Russian, Italian, French, Korean, Japanese, German, Portuguese

10,000 characters

Validity: Valid for 90 days after you activate Model Studio

Billing is based on the number of input characters. The calculation rules are as follows:

Each Chinese character, including simplified and traditional Chinese, Japanese Kanji, and Korean Hanja, counts as 2 characters.
Any other character, such as an English letter, a punctuation mark, or a space, counts as 1 character.

Qwen3-TTS-VC-Realtime

Model

Version

Price

Supported languages

Free quota (Note)

qwen3-tts-vc-realtime-2025-11-27

Snapshot

$0.13/10,000 characters

Chinese, English, Spanish, Russian, Italian, French, Korean, Japanese, German, Portuguese

10,000 characters

Valid for 90 days after you activate Model Studio.

Billing is based on the number of input characters. The calculation rules are as follows:

Each Chinese character, including simplified and traditional Chinese, Japanese Kanji, and Korean Hanja, counts as 2 characters.
Any other character, such as an English letter, a punctuation mark, or a space, counts as 1 character.

Qwen3-TTS-Flash-Realtime

Model	Version	Price	Supported languages	Free quota (Note)
qwen3-tts-flash-realtime This model has the same capabilities as qwen3-tts-flash-realtime-2025-09-18.	Stable	$0.13/10,000 characters	Chinese (Mandarin, Beijing, Shanghai, Sichuan, Nanjing, Shaanxi, Minnan, Tianjin, Cantonese), English, Spanish, Russian, Italian, French, Korean, Japanese, German, Portuguese	If you activate Model Studio before 00:00 on November 13, 2025: 2,000 characters If you activate Model Studio on or after 00:00 on November 13, 2025: 10,000 characters Valid for 90 days after you activate Model Studio
qwen3-tts-flash-realtime-2025-11-27	Snapshot			10,000 characters Valid for 90 days after you activate Model Studio
qwen3-tts-flash-realtime-2025-09-18	Snapshot			If you activate Model Studio before 00:00 on November 13, 2025: 2,000 characters If you activate Model Studio on or after 00:00 on November 13, 2025: 10,000 characters Valid for 90 days after you activate Model Studio

Billing is based on the number of input characters. The calculation rules are as follows:

Each Chinese character, including simplified and traditional Chinese, Japanese Kanji, and Korean Hanja, counts as 2 characters.
Any other character, such as an English letter, a punctuation mark, or a space, counts as 1 character.

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the China (Beijing) region, and inference computing resources are restricted to Mainland China.

Qwen3-TTS-VD-Realtime

Model	Version	Price	Supported languages	Free quota (Note)
qwen3-tts-vd-realtime-2025-12-16	Snapshot	$0.143353 per 10,000 characters	Chinese, English, Spanish, Russian, Italian, French, Korean, Japanese, German, Portuguese	No free quota

Billing is based on the number of input characters. The calculation rules are as follows:

Each Chinese character, including simplified and traditional Chinese, Japanese Kanji, and Korean Hanja, counts as 2 characters.
Any other character, such as an English letter, a punctuation mark, or a space, counts as 1 character.

Qwen3-TTS-VC-Realtime

Model	Version	Price	Supported languages	Free quota (Note)
qwen3-tts-vc-realtime-2025-11-27	Snapshot	$0.143353/10,000 characters	Chinese, English, Spanish, Russian, Italian, French, Korean, Japanese, German, Portuguese	No free quota

Billing is based on the number of input characters. The calculation rules are as follows:

Each Chinese character, including simplified and traditional Chinese, Japanese Kanji, and Korean Hanja, counts as 2 characters.
Any other character, such as an English letter, a punctuation mark, or a space, counts as 1 character.

Qwen3-TTS-Flash-Realtime

Model	Version	Price	Supported languages	Free quota (Note)
qwen3-tts-flash-realtime Functionally identical to qwen3-tts-flash-realtime-2025-09-18.	Stable	$0.143353/10,000 characters	Chinese (Mandarin, Beijing, Shanghai, Sichuan, Nanjing, Shaanxi, Minnan, Tianjin, Cantonese), English, Spanish, Russian, Italian, French, Korean, Japanese, German, Portuguese	No free quota
qwen3-tts-flash-realtime-2025-11-27	Snapshot
qwen3-tts-flash-realtime-2025-09-18	Snapshot

Billing is based on the number of input characters. The calculation rules are as follows:

Each Chinese character, including simplified and traditional Chinese, Japanese Kanji, and Korean Hanja, counts as 2 characters.
Any other character, such as an English letter, a punctuation mark, or a space, counts as 1 character.

Qwen-TTS-Realtime

Model	Version	Context window	Max input	Max output	Input cost	Output cost	Supported languages	Free quota (Note)
		(Tokens)			(per 1,000 tokens)
qwen-tts-realtime This model has the same capabilities as qwen-tts-realtime-2025-07-15.	Stable	8,192	512	7,680	$0.345	$1.721	Chinese, English	No free quota
qwen-tts-realtime-latest This model has the same capabilities as qwen-tts-realtime-2025-07-15.	Latest						Chinese, English
qwen-tts-realtime-2025-07-15	Snapshot						Chinese, English

Audio is converted to tokens at a rate of 50 tokens per second. Audio clips shorter than 1 second are billed as 50 tokens.

Qwen voice cloning

Voice cloning uses a model for feature extraction to clone voices without training. You can provide as little as 10 to 20 seconds of audio to generate a highly similar and natural-sounding custom voice. Usage | API reference

International

Model

Price

Free quota (Note)

qwen-voice-enrollment

$0.01/voice

1,000 voices

Valid for 90 days after you activate Model Studio

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the China (Beijing) region, and inference computing resources are restricted to Mainland China.

Model	Price	Free quota (Note)
qwen-voice-enrollment	$0.01/voice	No free quota

Qwen voice design

Voice design generates custom voices from text descriptions. It supports multilingual and multi-dimensional voice feature definitions. This feature is suitable for various applications, such as ad dubbing, character creation, and audio content creation. Usage | API reference

International

Model

Price

Free quota (Note)

qwen-voice-design

$0.20 per voice

10 timbres

Valid for 90 days after you activate Model Studio

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the China (Beijing) region, and inference computing resources are restricted to Mainland China.

Model	Price	Free quota (Note)
qwen-voice-design	$0.20 per voice	No free quota

CosyVoice speech synthesis

CosyVoice is a next-generation generative speech synthesis model from Tongyi Lab. Built on large-scale pre-trained language models, CosyVoice deeply integrates text understanding with speech generation and supports real-time, streaming text-to-speech synthesis. Usage | API reference

Note

Model	Price	Free quota (Note)
cosyvoice-v3-plus	$0.286706 per 10,000 characters	No free quota
cosyvoice-v3-flash	$0.14335 per 10,000 characters
cosyvoice-v2	$0.286706 per 10,000 characters

Characters are calculated as follows: Each Chinese character, including simplified and traditional Chinese, Japanese Kanji, and Korean Hanja, counts as 2 characters. Any other character, such as letters, numbers, Japanese Kana, and Korean Hangul, counts as 1 character. Content within SSML tags is not billed.

Speech recognition (speech-to-text) and translation (speech-to-translation)

Qwen3-LiveTranslate-Flash

Qwen3-LiveTranslate-Flash-Realtime

Qwen3-LiveTranslate-Flash-Realtime is a multilingual model for real-time audio and video translation. It recognizes 18 languages and provides real-time audio translations in 10 languages.

Core features:

Multilingual support: Supports 18 languages and 6 Chinese dialects, including Chinese, English, French, German, Russian, Japanese, and Korean. It also supports dialects such as Mandarin, Cantonese, and Sichuanese.
Vision enhancement: Uses visual content to improve translation accuracy. The model analyzes visual cues, such as lip movements, actions, and on-screen text, to enhance translation accuracy in noisy environments or when speech is ambiguous.
3-second latency: Achieves a simultaneous interpretation latency of as low as 3 seconds.
Lossless simultaneous interpretation: Resolves cross-lingual word order issues using semantic unit prediction technology. The quality of real-time translation is comparable to that of offline translation.
Natural voice: Generates speech with a natural, human-like voice. The model automatically adjusts its tone and emotion based on the source audio content.

Usage | API reference

International

Model	Version	Context window	Max input	Max output	Free quota (Note)
		(Tokens)
qwen3-livetranslate-flash-realtime This model has the same capabilities as qwen3-livetranslate-flash-realtime-2025-09-22.	Stable	53,248	49,152	4,096	1 million tokens for each version Valid for 90 days after you activate Model Studio
qwen3-livetranslate-flash-realtime-2025-09-22	Snapshot

After the free quota is used, billing for input and output is calculated as follows:

Billing Item	Price (per million tokens)
Input: Audio	$10
Input: Image	$1.3

Billing Items for Output	Price (per million tokens)
Text	$10
Audio	$38

Token calculation:

Audio: Each second of input or output audio consumes 12.5 tokens.
Image: Each 28 × 28 pixel input consumes 0.5 tokens.

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the China (Beijing) region, and inference computing resources are restricted to Mainland China.

Model	Version	Context window	Max input	Max output	Free quota (Note)
		(Tokens)
qwen3-livetranslate-flash-realtime This model has the same capabilities as qwen3-livetranslate-flash-realtime-2025-09-22.	Stable	53,248	49,152	4,096	No free quota
qwen3-livetranslate-flash-realtime-2025-09-22	Snapshot

Billing for input and output is calculated as follows:

Billing Item Input	Price (per million tokens)
Input: Audio	$9.175
Input image	$1.147

Billable Items for Output	Price (per million tokens)
Text	$9.175
Audio	$34.405

Token calculation:

Audio: Each second of input or output audio consumes 12.5 tokens.
Image: Each 28 × 28 pixel input consumes 0.5 tokens.

Qwen audio file recognition

Based on the Qwen multimodal foundation model, this feature supports multilingual recognition, singing recognition, and noise rejection. Usage | API reference

International

Qwen3-ASR-Flash-Filetrans

Model

Version

Supported languages

Supported sample rates

Price

Free quota (Note)

qwen3-asr-flash-filetrans

Provides the same capabilities as qwen3-asr-flash-filetrans-2025-11-17.

Stable

Chinese (Mandarin, Sichuanese, Minnan, Wu, and Cantonese), English, Japanese, German, Korean, Russian, French, Portuguese, Arabic, Italian, Spanish, Hindi, Indonesian, Thai, Turkish, Ukrainian, Vietnamese, Czech, Danish, Filipino, Finnish, Icelandic, Malay, Norwegian, Polish, and Swedish

Any

$0.000035/second

36,000 seconds (10 hours)

Valid for 90 days after you activate Model Studio

qwen3-asr-flash-filetrans-2025-11-17

Snapshot

Qwen3-ASR-Flash

Model

Version

Supported languages

Supported sample rates

Unit price

Free quota (Note)

qwen3-asr-flash

Same capabilities as qwen3-asr-flash-2025-09-08.

Stable

Chinese (Mandarin, Sichuanese, Minnan, Wu, Cantonese), English, Japanese, German, Korean, Russian, French, Portuguese, Arabic, Italian, Spanish, Hindi, Indonesian, Thai, Turkish, Ukrainian, Vietnamese, Czech, Danish, Filipino, Finnish, Icelandic, Malay, Norwegian, Polish, and Swedish

Any

$0.000035/second

36,000 seconds (10 hours)

Validity: Valid for 90 days after you activate Model Studio

qwen3-asr-flash-2025-09-08

Snapshot

US

In the US deployment mode, the endpoint and data storage are located in the US (Virginia) region, and inference computing resources are restricted to the United States.

Model

Version

Supported languages

Supported sample rates

Price

Free quota (Note)

qwen3-asr-flash-us

This model provides the same capabilities as qwen3-asr-flash-2025-09-08-us.

Stable

Any

$0.000035/second

No free quota

qwen3-asr-flash-2025-09-08-us

Snapshot

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the China (Beijing) region, and inference computing resources are restricted to Mainland China.

Qwen3-ASR-Flash-Filetrans

Model

Version

Supported languages

Supported sample rates

Price

Free quota (Note)

qwen3-asr-flash-filetrans

Current equivalent: qwen3-asr-flash-filetrans-2025-11-17

Stable

Any

$0.000032/second

No free quota

qwen3-asr-flash-filetrans-2025-11-17

Snapshot

Qwen3-ASR-Flash

Model

Version

Supported languages

Supported sample rates

Price

Free quota (Note)

qwen3-asr-flash

Functionally identical to qwen3-asr-flash-2025-09-08.

Stable

Any

$0.000032/second

No free quota

qwen3-asr-flash-2025-09-08

Snapshot

Qwen real-time speech recognition

The Qwen real-time speech recognition model provides automatic language detection. It detects 11 types of speech and accurately transcribes audio in complex environments. Usage | API reference

International

Model

Version

Supported languages

Supported sample rates

Price

Free quota (Note)

qwen3-asr-flash-realtime

Same capabilities as qwen3-asr-flash-realtime-2025-10-27.

Stable

8 kHz, 16 kHz

$0.00009/second

36,000 seconds (10 hours)

Valid for 90 days after you activate Model Studio

qwen3-asr-flash-realtime-2025-10-27

Snapshot

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the China (Beijing) region, and inference computing resources are restricted to Mainland China.

Model

Version

Supported languages

Supported sample rates

Price

Free quota (Note)

qwen3-asr-flash-realtime

This model has the same capabilities as qwen3-asr-flash-realtime-2025-10-27.

Stable

8 kHz, 16 kHz

$0.000047/second

No free quota

qwen3-asr-flash-realtime-2025-10-27

Snapshot

Paraformer speech recognition

Paraformer is a speech recognition model from Tongyi Lab. It is available in two versions: audio file recognition and real-time speech recognition.

Audio file recognition

Usage | API reference

Note

Model	Supported languages	Supported sample rates	Scenarios	Supported audio formats	Price	Free quota (Note)
paraformer-v2	Chinese (Mandarin, Cantonese, Wu, Minnan, Northeastern, Gansu, Guizhou, Henan, Hubei, Hunan, Ningxia, Shanxi, Shaanxi, Shandong, Sichuan, Tianjin, Jiangxi, Yunnan, Shanghai), English, Japanese, Korean, German, French, Russian	Any	Live stream	aac, amr, avi, flac, flv, m4a, mkv, mov, mp3, mp4, mpeg, ogg, opus, wav, webm, wma, wmv	$0.000012/second	No free quota
paraformer-8k-v2	Chinese (Mandarin)	8 kHz	Phone calls		$0.000012/second	No free quota

Real-time speech recognition

Usage | API reference

Note

Model	Supported languages	Supported sample rates	Scenarios	Supported audio formats	Price	Free quota (Note)
paraformer-realtime-v2	Chinese (Mandarin, Cantonese, Wu, Minnan, Northeastern, Gansu, Guizhou, Henan, Hubei, Hunan, Ningxia, Shanxi, Shaanxi, Shandong, Sichuan, Tianjin, Jiangxi, Yunnan, and Shanghai), English, Japanese, Korean, German, French, and Russian You can switch between multiple languages.	Any	Live video streaming and conferences	pcm, wav, mp3, opus, speex, aac, amr	$0.000035/second	No free quota
paraformer-realtime-8k-v2	Chinese (Mandarin)	8 kHz	Call centers and more	pcm, wav, mp3, opus, speex, aac, amr	$0.000035/second	No free quota

Fun-ASR speech recognition

Fun-ASR is a speech recognition model from the Tongyi Fun series. It is available in two versions: audio file recognition and real-time speech recognition.

Audio file recognition

Usage | API reference

International

Model	Version	Supported languages	Supported sample rates	Scenarios	Supported audio formats	Price	Free quota (Note)
fun-asr Its capabilities are the same as fun-asr-2025-11-07.	Stable	Chinese (Mandarin, Cantonese, Wu, Minnan, Hakka, Gan, Xiang, and Jin), Mandarin accents from regions such as Zhongyuan, Southwest, Ji-Lu, Jianghuai, Lan-Yin, Jiao-Liao, Northeast, Beijing, Hong Kong, and Taiwan (including accents from Henan, Shaanxi, Hubei, Sichuan, Chongqing, Yunnan, Guizhou, Guangdong, Guangxi, Hebei, Tianjin, Shandong, Anhui, Nanjing, Jiangsu, Hangzhou, Gansu, and Ningxia), English, and Japanese	Any	Live stream, phone calls, conference interpretation, and more	aac, amr, avi, flac, flv, m4a, mkv, mov, mp3, mp4, mpeg, ogg, opus, wav, webm, wma, and wmv	$0.000035/second	36,000 seconds (10 hours) Valid for 90 days
fun-asr-2025-11-07 Compared to fun-asr-2025-08-25, this version is optimized for far-field VAD to improve recognition accuracy.	Snapshot
fun-asr-2025-08-25		Chinese (Mandarin), English
fun-asr-mtl Its capabilities are the same as fun-asr-mtl-2025-08-25.	Stable	Chinese (Mandarin and Cantonese), English, Japanese, Korean, Vietnamese, Indonesian, Thai, Malay, Filipino, Arabic, Hindi, Bulgarian, Croatian, Czech, Danish, Dutch, Estonian, Finnish, Greek, Hungarian, Irish, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, and Swedish
fun-asr-mtl-2025-08-25	Snapshot

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the China (Beijing) region, and inference computing resources are restricted to Mainland China.

Model	Version	Supported languages	Supported sample rates	Scenarios	Supported audio formats	Price	Free quota (Note)
fun-asr Same capabilities as fun-asr-2025-11-07.	Stable	Chinese (Mandarin, Cantonese, Wu, Minnan, Hakka, Gan, Xiang, and Jin), Mandarin accents from regions such as Zhongyuan, Southwest, Ji-Lu, Jianghuai, Lan-Yin, Jiao-Liao, Northeast, Beijing, Hong Kong, and Taiwan (including accents from Henan, Shaanxi, Hubei, Sichuan, Chongqing, Yunnan, Guizhou, Guangdong, Guangxi, Hebei, Tianjin, Shandong, Anhui, Nanjing, Jiangsu, Hangzhou, Gansu, and Ningxia), English, and Japanese	Any	Live stream, phone calls, conference interpretation, and more	aac, amr, avi, flac, flv, m4a, mkv, mov, mp3, mp4, mpeg, ogg, opus, wav, webm, wma, and wmv	$0.000032/second	No free quota
fun-asr-2025-11-07 Compared to fun-asr-2025-08-25, this version is optimized for far-field VAD to improve recognition accuracy.	Snapshot
fun-asr-2025-08-25		Chinese (Mandarin), English
fun-asr-mtl Same capabilities as fun-asr-mtl-2025-08-25.	Stable	Chinese (Mandarin and Cantonese), English, Japanese, Korean, Vietnamese, Indonesian, Thai, Malay, Filipino, Arabic, Hindi, Bulgarian, Croatian, Czech, Danish, Dutch, Estonian, Finnish, Greek, Hungarian, Irish, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, and Swedish
fun-asr-mtl-2025-08-25	Snapshot

Real-time speech recognition

Usage | API reference

International

Model

Version

Supported languages

Supported sample rates

Scenarios

Supported audio formats

Price

Free quota (Note)

fun-asr-realtime

The capabilities of this model are the same as fun-asr-realtime-2025-11-07.

Stable

Chinese (Mandarin, Cantonese, Wu, Minnan, Hakka, Gan, Xiang, and Jin), English, and Japanese. This model also supports Mandarin accents from regions such as Zhongyuan, Southwest, Ji-Lu, Jianghuai, Lan-Yin, Jiao-Liao, Northeast, Beijing, Hong Kong, and Taiwan. Additionally, it supports accents from areas such as Henan, Shaanxi, Hubei, Sichuan, Chongqing, Yunnan, Guizhou, Guangdong, Guangxi, Hebei, Tianjin, Shandong, Anhui, Nanjing, Jiangsu, Hangzhou, Gansu, and Ningxia.

16 kHz

Live video streaming, video conferencing, call centers, and more

pcm, wav, mp3, opus, speex, aac, and amr

$0.00009/second

36,000 seconds (10 hours)

Valid for 90 days

fun-asr-realtime-2025-11-07

Snapshot

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the China (Beijing) region, and inference computing resources are restricted to Mainland China.

Model	Version	Supported languages	Supported sample rates	Scenarios	Supported audio formats	Price	Free quota (Note)
fun-asr-realtime This model has the same capabilities as fun-asr-realtime-2025-11-07.	Stable	Chinese (Mandarin, Cantonese, Wu, Minnan, Hakka, Gan, Xiang, and Jin), English, and Japanese. This model also supports Mandarin accents from regions such as Zhongyuan, Southwest, Ji-Lu, Jianghuai, Lan-Yin, Jiao-Liao, Northeast, Beijing, Hong Kong, and Taiwan. Additionally, it supports accents from areas such as Henan, Shaanxi, Hubei, Sichuan, Chongqing, Yunnan, Guizhou, Guangdong, Guangxi, Hebei, Tianjin, Shandong, Anhui, Nanjing, Jiangsu, Hangzhou, Gansu, and Ningxia.	16 kHz	Live video streaming, video conferences, call centers, and more	pcm, wav, mp3, opus, speex, aac, and amr	$0.000047/second	No free quota
fun-asr-realtime-2025-11-07 Compared to fun-asr-realtime-2025-09-15, this version is optimized for far-field VAD to improve recognition accuracy.	Snapshot
fun-asr-realtime-2025-09-15		Chinese (Mandarin), English

Text embedding

Text embedding models convert text into numerical representations for tasks such as search, clustering, recommendation, and classification. Billing for these models is based on the number of input tokens. API reference

International

In the international deployment mode, endpoints and data storage are located in the Singapore region. Inference computing resources are scheduled globally, excluding Mainland China.

Model

Embedding dimensions

Batch size

Max tokens per batch (Note)

Supported languages

Price

(Million input tokens)

Free quota

(Note)

text-embedding-v4

Part of the Qwen3-Embedding series

2,048, 1,536, 1,024 (default), 768, 512, 256, 128, or 64

8,192

More than 100 major languages, such as Chinese, English, Spanish, French, Portuguese, Indonesian, Japanese, Korean, German, and Russian, and various programming languages

$0.07

1 million tokens

Valid for 90 days after you activate Model Studio.

text-embedding-v3

1,024 (default), 768, or 512

8,192

Over 50 languages, such as Chinese, English, Spanish, French, Portuguese, Indonesian, Japanese, Korean, German, and Russian

500,000 tokens

Valid for 90 days after you activate Model Studio.

Mainland China

In the Mainland China deployment mode, endpoints and data storage are located in the Beijing region. Inference computing resources are limited to Mainland China.

Model

Embedding dimensions

Batch size

Max tokens per batch (Note)

Supported languages

Price

(Million input tokens)

Free quota

(Note)

text-embedding-v4

Part of the Qwen3-Embedding series

2,048, 1,536, 1,024 (default), 768, 512, 256, 128, or 64

8,192

More than 100 major languages, such as Chinese, English, Spanish, French, Portuguese, Indonesian, Japanese, Korean, German, and Russian, and various programming languages

$0.072

No free quota

Note

Batch size is the max number of texts that a single API call can process. For example, the batch size for text-embedding-v4 is 10. This means a single request can vectorize up to 10 texts, and each text cannot exceed 8,192 tokens. This limit applies to:

String array input: The array can contain up to 10 elements.
File input: The text file can contain up to 10 lines of text.

Multimodal embedding

A multimodal embedding model converts text, images, and videos into a vector of floating-point numbers. The model is suitable for applications such as video classification, image classification, and image-text retrieval. API reference

International

In the international deployment mode, the endpoint and data storage are located in the Singapore region. Inference computing resources are scheduled globally, excluding Mainland China.

Model

Data type

Embedding dimensions

Unit price (Million input tokens)

Free quota (Note)

tongyi-embedding-vision-plus

float(32)

1,152

$0.09

1 million tokens

Valid for 90 days after you activate Model Studio.

tongyi-embedding-vision-flash

float(32)

768

Image/Video: $0.03

Text: $0.09

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the Beijing region. Inference computing resources are limited to Mainland China.

Model	Data type	Embedding dimensions	Price (1,000 input tokens)	Free quota (Note)
multimodal-embedding-v1	float(32)	1,024	Free trial	No token limit

Text rerank

This feature is typically used for semantic retrieval. Given a query, it sorts a list of candidate documents in descending order of their semantic relevance. API reference

International

Model

Max number of documents

Max input tokens per item

Max Input Tokens

Supported languages

Price (Million input tokens)

qwen3-rerank

500

4,000

30,000

Over 100 major languages, such as Chinese, English, Spanish, French, Portuguese, Indonesian, Japanese, Korean, German, and Russian

$0.1

Max input tokens per item: Each query or document is limited to 4,000 tokens. Input that exceeds this limit is truncated.
Max number of documents: Each request is limited to 500 documents.
Max input tokens: The total number of tokens for all queries and documents in a single request is limited to 30,000.

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the Beijing region. Inference computing resources are limited to Mainland China.

Model	Max number of documents	Max input tokens per item	Max input tokens	Supported languages	Price (Million input tokens)
gte-rerank-v2	500	4,000	30,000	More than 50 languages, such as Chinese, English, Japanese, Korean, Thai, Spanish, French, Portuguese, German, Indonesian, and Arabic	$0.115

Max input tokens per item: Each query or document is limited to 4,000 tokens. Input that exceeds this limit is truncated.
Max number of documents: Each request is limited to 500 documents.
Max input tokens: The total number of tokens for all queries and documents in a single request is limited to 30,000.

Domain specific

Intent recognition

The Qwen intent recognition model can quickly and accurately parse user intents in milliseconds and select the appropriate tools to resolve user issues. API reference | Usage

Note

Model	Context window	Max input	Max output	Input cost	Output cost
Model	(Tokens)			(Million tokens)
tongyi-intent-detect-v3	8,192	8,192	1,024	$0.058	$0.144

Role playing

Qwen's role-playing model is ideal for scenarios that require human-like conversation, such as virtual social interactions, NPCs in games, replicating IP characters, hardware, toys, and in-vehicle systems. Compared to other Qwen models, this model offers enhanced capabilities in character fidelity, conversation progression, and empathetic listening. Usage

International

Model	Context window	Max input	Max output	Input cost	Output cost
Model	(Tokens)			(Million tokens)
qwen-plus-character-ja	8,192	7,680	512	$0.5	$1.4

Mainland China

In the Mainland China deployment mode, the endpoint and data storage are located in the Beijing region. Inference computing resources are limited to Mainland China.

Model	Context window	Max input	Max output	Input cost	Output cost
Model	(Tokens)			(Million tokens)
qwen-plus-character	32,768	32,000	4,096	$0.115	$0.287

Retired models

Retired on August 20, 2025

Qwen2

The open-source Qwen2 model from Alibaba Cloud. Usage | API reference | Try it online

Model	Context window	Max input	Max output	Input cost	Output cost	Alternative models
	(Tokens)			(Million tokens)
qwen2-72b-instruct	131,072	128,000	6,144	Free for a limited time		Qwen3, DeepSeek, Kimi, and others
qwen2-57b-a14b-instruct	65,536	63,488
qwen2-7b-instruct	131,072	128,000

Qwen1.5

The open-source Qwen1.5 model from Alibaba Cloud. Usage | API reference | Try it online

Model	Context window	Max input	Max output	Input cost	Output cost	Alternative models
	(Tokens)			(Million tokens)
qwen1.5-110b-chat	8,000	6,000	2,000	Free for a limited time		Qwen3, DeepSeek, Kimi, and others
qwen1.5-72b-chat
qwen1.5-32b-chat
qwen1.5-14b-chat
qwen1.5-7b-chat