AI inference fees - MaxCompute - Alibaba Cloud Documentation Center

MaxCompute offers the AI inference, a new, out-of-the-box, pay-as-you-go feature for data processing and offline inference with large models. This topic describes the billing rules.

Overview

The MaxCompute model inference service provides large model inference capabilities, allowing you to call ready-to-use models directly from your SQL and MaxFrame jobs by using a built-in AI function. You are charged based on the number of input and output tokens used.

Supported regions: The model inference service is available only in the following regions.
- Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)
- Other regions: Singapore

Supported models: To use the following models, you must first enable the model inference service in a supported region, such as China (Beijing).

Model	Supported input	Description
`qwen3.7-max`	Text input only	The flagship model of the Qwen 3.7 series. It offers comprehensive upgrades in inference, code generation, and multilingual understanding, making it suitable for highly complex tasks.
`qwen3.7-plus`	Multimodal input supported	A balanced model in the Qwen 3.7 series that balances performance and cost, suitable for enterprise-level scenarios such as long-text analysis and multi-turn conversations.
`qwen3-vl-embedding`	Multimodal input supported	A Qwen multimodal embedding model. It creates vector representations from mixed image-text inputs and is suitable for cross-modal retrieval and image-text matching.
`text-embedding-v4`	Text input only	A high-precision text embedding model suitable for tasks like semantic search, clustering, and similarity calculation.
`qwen3.6-plus`	Multimodal input supported	A balanced model in the Qwen 3.6 series that balances performance and cost, suitable for general-purpose tasks such as conversation, summarization, and analysis.
`qwen3.6-flash`	Multimodal input supported	A lightweight and high-speed model from the Qwen 3.6 series. Its fast response times and low costs make it ideal for high-concurrency, low-latency online services.
`deepseek-v4-pro`	Text input only	The fourth-generation flagship inference model from DeepSeek. It excels at difficult tasks such as mathematics, coding, and complex logical reasoning.
`deepseek-v4-flash`	Text input only	A fourth-generation lightweight model from DeepSeek. With fast, cost-effective inference, it is ideal for daily conversation and lightweight reasoning tasks.
`qwen3.5-397b-a17b`	Multimodal input supported	A Mixture-of-Experts (MoE) model from the Qwen 3.5 series. It features efficient parameter activation, a vast knowledge base, and multi-step reasoning capabilities. It is designed for challenging logical deduction, full-stack code generation, and in-depth knowledge Q&A.
`qwen3-asr-flash`	Multimodal input supported	A lightweight automatic speech recognition (ASR) model from the Qwen 3 series. Its low-latency, high-concurrency real-time transcription is ideal for audio processing scenarios such as meeting transcription, live subtitling, and voice-command recognition.
`qwen3-max` (To be deprecated)	Text input only	A high-performance large language model from the Qwen series, suitable for complex inference and content generation.

Note

Currently, model inference service fees are incurred only when you use the public models listed above through a built-in AI function in MaxFrame and SQL jobs. You are billed only for successful jobs.

Billing rules

The AI inference is billed by token usage, based on the following factors:

Region
ModelName
TokenType: The token type (input or output) and its corresponding usage tier.

Pricing

A single SQL/MaxFrame job may involve multiple model calls. For models that use pricing tiers, the system independently counts the input tokens for each model call to determine the billing tier (also known as the token type). Usage for each tier is then tracked and billed separately.

Important

When deploying to the China region, compute resources for model inference are restricted to mainland China, and static data is stored in the selected region.
When deploying to the international region, compute resources for model inference are dynamically scheduled across the globe (excluding mainland China), and static data is stored in the selected region.

qwen3.7-max

Chinese mainland

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Input tokens	Scenario	Token type	Unit price
0 < tokens ≤ 1,048,576	model input	`input_token_tier1`	1.98 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.396 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.198 USD
	model input (creating explicit cache)	`input_token_tier1_create_cache`	2.475 USD
	model output (non-thinking)	`output_token_tier1`	5.9412 USD
	model output (thinking mode)	`output_token_tier1_thinking`	5.9412 USD

International

For deployments in the International region, compute resources for model inference are dynamically scheduled across global locations (excluding mainland China), and static data is stored in the selected region.

Supported region: Singapore

Input tokens	Scenario	Token type	Unit price
0 < tokens ≤ 1,048,576	model input	`input_token_tier1`	3 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.6 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.3 USD
	model input (creating explicit cache)	`input_token_tier1_create_cache`	3.75 USD
	model output (non-thinking)	`output_token_tier1`	9 USD
	model output (thinking mode)	`output_token_tier1_thinking`	9 USD

qwen3.7-plus

Chinese mainland

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Input tokens	Scenario	Token type	Price
0 < tokens ≤ 262,144	model input	`input_token_tier1`	0.3312 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.06624 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.03312 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	0.414 USD
	model output (non-thinking)	`output_token_tier1`	1.3212 USD
	model output (thinking mode)	`output_token_tier1_thinking`	1.3212 USD
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	0.9912 USD
	model input (implicit cache hit)	`input_token_tier2_cached`	0.19824 USD
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.09912 USD
	model input (create explicit cache)	`input_token_tier2_create_cache`	1.239 USD
	model output (non-thinking)	`output_token_tier2`	3.9612 USD
	model output (thinking mode)	`output_token_tier2_thinking`	3.9612 USD

International

Supported region: Singapore

Input tokens	Scenario	Token type	Price
0 < tokens ≤ 262,144	model input	`input_token_tier1`	0.48 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.096 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.048 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	0.6 USD
	model output (non-thinking)	`output_token_tier1`	1.92 USD
	model output (thinking mode)	`output_token_tier1_thinking`	1.92 USD
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	1.44 USD
	model input (implicit cache hit)	`input_token_tier2_cached`	0.288 USD
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.144 USD
	model input (create explicit cache)	`input_token_tier2_create_cache`	1.8 USD
	model output (non-thinking)	`output_token_tier2`	5.76 USD
	model output (thinking mode)	`output_token_tier2_thinking`	5.76 USD

qwen3-vl-embedding

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Price tier	Use case	Token type	Unit price
Text	model input	`input_token_tier1_text`	0.12 USD
Image & video	model input	`input_token_tier1_image`	0.3096 USD

text-embedding-v4

The text-embedding-v4 model is billed for model input tokens only; model output is free. Pricing is flat-rate, with all usage billed as input_token_tier1.

Chinese mainland

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Tier	Scenario	Token type	Price
no tiers	model input	`input_token_tier1`	0.0864 USD

International

Supported region: Singapore

Tier	Scenario	Token type	Price
No tiers	model input	`input_token_tier1`	0.084 USD

qwen3.6-plus

Chinese mainland

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Input token count	Scenario	Token type	Price / million tokens
0 < tokens ≤ 262,144	model input	`input_token_tier1`	0.3312 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.06624 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.03312 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	0.414 USD
	model output (non-thinking)	`output_token_tier1`	1.9812 USD
	model output (thinking mode)	`output_token_tier1_thinking`	1.9812 USD
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	1.3212 USD
	model input (implicit cache hit)	`input_token_tier2_cached`	0.26424 USD
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.13212 USD
	model input (create explicit cache)	`input_token_tier2_create_cache`	1.6515 USD
	model output (non-thinking)	`output_token_tier2`	7.9224 USD
	model output (thinking mode)	`output_token_tier2_thinking`	7.9224 USD

International

Supported region: Singapore

Input token count	Scenario	Token type	Price / million tokens
0 < tokens ≤ 262,144	model input	`input_token_tier1`	0.6 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.12 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.06 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	0.75 USD
	model output (non-thinking)	`output_token_tier1`	3.6 USD
	model output (thinking mode)	`output_token_tier1_thinking`	3.6 USD
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	2.4 USD
	model input (implicit cache hit)	`input_token_tier2_cached`	0.48 USD
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.24 USD
	model input (create explicit cache)	`input_token_tier2_create_cache`	3 USD
	model output (non-thinking)	`output_token_tier2`	7.2 USD
	model output (thinking mode)	`output_token_tier2_thinking`	7.2 USD

qwen3.6-flash

Chinese mainland

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Input tokens	Scenario	Token type	Price / million tokens
0 < tokens ≤ 262,144	model input	`input_token_tier1`	0.198 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.0396 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.0198 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	0.2475 USD
	model output (non-thinking)	`output_token_tier1`	1.188 USD
	model output (thinking mode)	`output_token_tier1_thinking`	1.188 USD
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	0.792 USD
	model input (implicit cache hit)	`input_token_tier2_cached`	0.1584 USD
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.0792 USD
	model input (create explicit cache)	`input_token_tier2_create_cache`	0.99 USD
	model output (non-thinking)	`output_token_tier2`	4.7532 USD
	model output (thinking mode)	`output_token_tier2_thinking`	4.7532 USD

International

Supported region: Singapore

Input tokens	Scenario	Token type	Price / million tokens
0 < tokens ≤ 262,144	model input	`input_token_tier1`	0.3 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.06 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.03 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	0.375 USD
	model output (non-thinking)	`output_token_tier1`	1.8 USD
	model output (thinking mode)	`output_token_tier1_thinking`	1.8 USD
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	1.2 USD
	model input (implicit cache hit)	`input_token_tier2_cached`	0.24 USD
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.12 USD
	model input (create explicit cache)	`input_token_tier2_create_cache`	1.5 USD
	model output (non-thinking)	`output_token_tier2`	4.8 USD
	model output (thinking mode)	`output_token_tier2_thinking`	4.8 USD

deepseek-v4-pro

Chinese mainland

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Tier range	Scenario	Token type	Price (per million tokens)
No tiers	model input	`input_token_tier1`	1.98 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.396 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.198 USD
	model input (explicit cache creation)	`input_token_tier1_create_cache`	2.475 USD
	model output (non-thinking)	`output_token_tier1`	3.9612 USD
	model output (thinking mode)	`output_token_tier1_thinking`	3.9612 USD

International

Supported regions: Singapore

Input tokens	Scenario	Token type	Price (per million tokens)
0 < tokens ≤ 1,048,576	model input	`input_token_tier1`	2.88 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.576 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.288 USD
	model input (explicit cache creation)	`input_token_tier1_create_cache`	3.6 USD
	model output (non-thinking)	`output_token_tier1`	5.76 USD
	model output (thinking mode)	`output_token_tier1_thinking`	5.76 USD

deepseek-v4-flash

Chinese mainland

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Pricing tier	Use case	Token type	Unit price
non-tiered	model input	`input_token_tier1`	0.1656 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.03312 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.01656 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	0.207 USD
	model output (non-thinking)	`output_token_tier1`	0.33 USD
	model output (thinking mode)	`output_token_tier1_thinking`	0.33 USD

International

Supported regions: Singapore

Pricing tier	Use case	Token type	Unit price
0 < tokens ≤ 1,048,576	model input	`input_token_tier1`	0.24 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.048 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.024 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	0.3 USD
	model output (non-thinking)	`output_token_tier1`	0.48 USD
	model output (thinking mode)	`output_token_tier1_thinking`	0.48 USD

qwen3.5-397b-a17b

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Input token count	Use case	Token type	Price (per 1M tokens)
0 < tokens ≤ 131,072	Model input	`input_token_tier1`	0.2064 USD
	Model input (implicit cache hit)	`input_token_tier1_cached`	0.04128 USD
	Model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.02064 USD
	Model input (create explicit cache)	`input_token_tier1_create_cache`	0.258 USD
	Model output (non-thinking)	`output_token_tier1`	1.2384 USD
	Model output (thinking mode)	`output_token_tier1_thinking`	1.2384 USD
131,072 < tokens ≤ 262,144	Model input	`input_token_tier2`	0.516 USD
	Model input (implicit cache hit)	`input_token_tier2_cached`	0.1032 USD
	Model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.0516 USD
	Model input (create explicit cache)	`input_token_tier2_create_cache`	0.645 USD
	Model output (non-thinking)	`output_token_tier2`	3.096 USD
	Model output (thinking mode)	`output_token_tier2_thinking`	3.096 USD

qwen3-asr-flash

For qwen3-asr-flash, charges apply only to model input tokens. This model uses a single pricing tier, billed as input_token_tier1.

Chinese mainland

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Tier	Use case	Token type	Unit price (per million tokens)
No tiers	model input	`input_token_tier1`	1.536 USD

International

Supported region: Singapore

Tier	Use case	Token type	Unit price (per million tokens)
No tiers	model input	`input_token_tier1`	1.68 USD

qwen3-max (To be deprecated)

Chinese mainland

For deployments in the China region, compute resources for model inference are limited to mainland China, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)

Input tokens	Scenario	Token type	Price
0 < tokens ≤ 32,768	model input	`input_token_tier1`	0.4308 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.08616 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.04308 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	0.5385 USD
	model output (non-thinking)	`output_token_tier1`	1.7208 USD
	model output (thinking mode)	`output_token_tier1_thinking`	1.7208 USD
32,768 < tokens ≤ 131,072	model input	`input_token_tier2`	0.6888 USD
	model input (implicit cache hit)	`input_token_tier2_cached`	0.13776 USD
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.06888 USD
	model input (create explicit cache)	`input_token_tier2_create_cache`	0.861 USD
	model output (non-thinking)	`output_token_tier2`	2.7528 USD
	model output (thinking mode)	`output_token_tier2_thinking`	2.7528 USD
131,072 < tokens ≤ 258,048	model input	`input_token_tier3`	1.2048 USD
	model input (implicit cache hit)	`input_token_tier3_cached`	0.24096 USD
	model input (explicit cache hit)	`input_token_tier3_cached_explicit`	0.12048 USD
	model input (create explicit cache)	`input_token_tier3_create_cache`	1.506 USD
	model output (non-thinking)	`output_token_tier3`	4.8168 USD
	model output (thinking mode)	`output_token_tier3_thinking`	4.8168 USD

International

Supported region: Singapore

Input tokens	Scenario	Token type	Price
0 < tokens ≤ 32,768	model input	`input_token_tier1`	1.44 USD
	model input (implicit cache hit)	`input_token_tier1_cached`	0.288 USD
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.144 USD
	model input (create explicit cache)	`input_token_tier1_create_cache`	1.8 USD
	model output (non-thinking)	`output_token_tier1`	7.2 USD
	model output (thinking mode)	`output_token_tier1_thinking`	7.2 USD
32,768 < tokens ≤ 131,072	model input	`input_token_tier2`	2.88 USD
	model input (implicit cache hit)	`input_token_tier2_cached`	0.576 USD
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.288 USD
	model input (create explicit cache)	`input_token_tier2_create_cache`	3.6 USD
	model output (non-thinking)	`output_token_tier2`	14.4 USD
	model output (thinking mode)	`output_token_tier2_thinking`	14.4 USD
131,072 < tokens ≤ 258,048	model input	`input_token_tier3`	3.6 USD
	model input (implicit cache hit)	`input_token_tier3_cached`	0.72 USD
	model input (explicit cache hit)	`input_token_tier3_cached_explicit`	0.36 USD
	model input (create explicit cache)	`input_token_tier3_create_cache`	4.5 USD
	model output (non-thinking)	`output_token_tier3`	18 USD
	model output (thinking mode)	`output_token_tier3_thinking`	18 USD

Billing

Billing frequency: Hourly
Due to data aggregation, billing for inference fees may be delayed for several hours after a job completes. The charges displayed in Billing Management are final.
Procedure
1. Log in to Billing Management.
2. In the left navigation pane, choose Billing > Bill Details.
3. On the Bill Details page, set Product Name to MaxCompute and Product Name to MaxCompute model computing service to view your detailed bill.

FAQ

Q: How do I estimate my inference cost?
A: You can estimate your inference cost using the pricing table and the expected lengths of your input and output. For example, if each call to the qwen3-max model averages 2,000 input tokens and 1,000 output tokens, and a job requires 1,000 inferences, the approximate list price for the job is CNY 18.
Q: Is a free tier or trial available?
A: This service is strictly pay-as-you-go and does not offer a free tier. We recommend starting with small-scale tests to validate the model's effectiveness and estimate costs.
Q: Can I set a usage limit to prevent cost overruns?
A: This service does not currently support setting a usage limit. Please monitor your billing to manage costs.

MaxCompute:AI inference fees

Overview

Billing rules

Pricing

qwen3.7-max

Chinese mainland

International

qwen3.7-plus

Chinese mainland

International

qwen3-vl-embedding

text-embedding-v4

Chinese mainland

International

qwen3.6-plus

Chinese mainland

International

qwen3.6-flash

Chinese mainland

International

deepseek-v4-pro

Chinese mainland

International

deepseek-v4-flash

Chinese mainland

International

qwen3.5-397b-a17b

qwen3-asr-flash

Chinese mainland

International

qwen3-max (To be deprecated)

Chinese mainland

International

Billing

FAQ

Related documentation