All Products
Search

This Product

    Alibaba Cloud Model Studio:Rate limits

    Document Center

    Alibaba Cloud Model Studio:Rate limits

    Last Updated:Feb 13, 2026

    To ensure fair use, Alibaba Cloud Model Studio applies basic rate limits. These limits are model-specific and linked to your Alibaba Cloud account. The limit is calculated based on the total calls to a model from all RAM users, workspaces, and API keys under your account. If you exceed the limit, API requests will fail. You must wait for the limit to reset before making another call.

    Rules

    • Account-level limits: Rate limits apply at the Alibaba Cloud account level. They are calculated based on the total calls from all RAM users, workspaces, and API keys under the account.

    • Model-specific limits: Each model has an independent rate limit. See the tables below for details.

    FAQ

    Why is rate limiting triggered?

    Check the error message:

    • Requests rate limit exceeded or You exceeded your current requests list: This error indicates that the call frequency limit was triggered.

    • Allocated quota exceeded or You exceeded your current quota: This error indicates that the token consumption limit was triggered.

    • Request rate increased too quickly: This error indicates that a sudden surge in call frequency triggered the system's stability protection, even if the Requests Per Minute (RPM) or Tokens Per Minute (TPM) limits were not reached.

    • For other errors, see Error messages to identify the cause.

    Note: In addition to RPM and TPM, rate limits may also be enforced at the per-second level. These limits are Requests per Second (RPS), calculated as RPM/60, and Requests per Second (RPS), calculated as TPM/60. A burst of requests in a short period can trigger rate limiting, even if the total number of calls is below the per-minute limit.

    How to view model call usage?

    One hour after you call a model, go to the Monitoring (Singapore or Beijing) page. Set the query conditions, such as the time range and workspace. Then, in the Models area, find the target model and click Monitor in the Actions column to view the model's call statistics. For more information, see the Monitoring document.

    Data is updated hourly. During peak periods, there may be an hour-level latency.

    image

    How long does it take to recover after a rate limit is triggered?

    The limit typically resets within one minute. If other errors occur, see Error messages for solutions.

    How to avoid rate limiting?

    1. Choose a model with a higher rate limit: Stable or latest versions have higher rate limits than older snapshot versions.

    2. Optimize your calling strategy

      • Adjust the call frequency: If you receive a "Requests rate limit exceeded" or "You exceeded your current requests list" error, reduce the call frequency.

      • Reduce token consumption: If you receive an "Allocated quota exceeded" or "You exceeded your current quota" error, shorten the input or output length.

      • Smooth the request rate: If a sudden increase in call frequency triggers system stability protection, you may receive a "Request rate increased too quickly" error. In this case, optimize your client-side calling logic. You can adopt a request smoothing strategy, such as uniform scheduling, exponential backoff, or a request queue buffer. This strategy distributes requests evenly over the time window and avoids instantaneous peaks.

    3. Add a backup model

      If you encounter a rate limit error, switch to a backup model to continue generation. This improves concurrency and reduces the failure rate. The following code shows an example of retrying a request with qwen-plus-2025-07-14 after a rate limit is triggered for qwen-plus-2025-07-28.

      Sample code

      import os
import asyncio
from openai import AsyncOpenAI, APIStatusError

# Configuration
API_KEY = os.getenv("DASHSCOPE_API_KEY")
# Primary model
MODEL = "qwen-plus-2025-07-28"
# Backup model
BACKUP_MODEL = "qwen-plus-2025-07-14"
# Test question
QUESTION = "Who are you?"
# Concurrency setting
NUM_REQUESTS = 10

client = AsyncOpenAI(
    api_key=API_KEY,
    base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1"
)

async def send_request(model):
    """Send a single request"""
    try:
        await client.chat.completions.create(
            model=model,
            messages=[{"role": "user", "content": QUESTION}]
        )
        return True
    except APIStatusError as e:
        if e.status_code == 429:
            print(f"[Rate limit triggered] Model {model}")
            return False
        raise
    except Exception as e:
        print(f"[Request failed] Model {model}, Error: {e}")
        return False

async def task(i):
    # Try the primary model
    if await send_request(MODEL):
        return True
    # If rate-limited, try the backup model
    return await send_request(BACKUP_MODEL)

async def main():
    results = await asyncio.gather(*(task(i) for i in range(NUM_REQUESTS)))
    print(f"Successful requests: {sum(results)}, Failed requests: {len(results) - sum(results)}")

if __name__ == "__main__":
    asyncio.run(main())

    4. Split tasks: Processing long conversations or large documents can consume many tokens quickly. Split large batch tasks into smaller batches and submit them at different times.

    5. Use batch inference: If you do not need real-time results, use batch inference (Batch API). It is not subject to real-time rate limits, but you must consider queuing and processing time.

    Text generation - Qwen

    Qwen language models

    Global

    In the Global deployment mode, the endpoint and data storage are located in the US (Virginia) region. Inference computing resources are dynamically scheduled worldwide.

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-max

    600

    1,000,000

    qwen3-max-2025-09-23

    60

    100,000

    qwen3-max-preview

    600

    1,000,000

    qwen-plus

    15,000

    5,000,000

    qwen-plus-2025-12-01

    60

    1,000,000

    qwen-plus-2025-09-11

    qwen-plus-2025-07-28

    qwen-flash

    15,000

    10,000,000

    qwen-flash-2025-07-28

    60

    1,000,000

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-max

    600

    1,000,000

    qwen3-max-2026-01-23

    qwen3-max-2025-09-23

    60

    100,000

    qwen3-max-preview

    600

    1,000,000

    qwen-max

    120

    100,000

    qwen-max-latest

    600

    1,000,000

    qwen-max-2025-01-25

    (qwen-max-0125)

    60

    100,000

    qwen-plus

    600

    1,000,000

    qwen-plus-latest

    600

    1,000,000

    qwen-plus-2025-12-01

    120

    1,000,000

    qwen-plus-2025-09-11

    120

    qwen-plus-2025-07-28

    60

    100,000

    qwen-plus-2025-07-14

    (qwen-plus-0714)

    qwen-plus-2025-04-28

    (qwen-plus-0428)

    1,000,000

    qwen-plus-2025-01-25

    (qwen-plus-0125)

    100,000

    qwen-flash

    600

    5,000,000

    qwen-flash-2025-07-28

    600

    5,000,000

    qwq-plus

    60

    100,000

    qwen-turbo

    240

    100,000

    qwen-turbo-latest

    600

    5,000,000

    qwen-turbo-2025-04-28

    (qwen-turbo-0428)

    60

    1,000,000

    qwen-turbo-2024-11-01

    (qwen-turbo-1101)

    5,000,000

    US

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-plus-us

    600

    1,000,000

    qwen-plus-2025-12-01-us

    60

    qwen-flash-us

    600

    5,000,000

    qwen-flash-2025-07-28-us

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-max

    30,000

    5,000,000

    qwen3-max-2026-01-23

    600

    1,000,000

    qwen3-max-2025-09-23

    60

    100,000

    qwen3-max-preview

    600

    1,000,000

    qwen-max

    1,200

    1,000,000

    qwen-max-latest

    qwen-max-2025-01-25

    (qwen-max-0125)

    60

    100,000

    qwen-max-2024-09-19

    (qwen-max-0919)

    qwen-plus

    30,000

    5,000,000

    qwen-plus-latest

    15,000

    1,200,000

    qwen-plus-2025-12-01

    120

    1,000,000

    qwen-plus-2025-09-11

    60

    qwen-plus-2025-07-28

    (qwen-plus-0728)

    qwen-plus-2025-07-14

    (qwen-plus-0714)

    100,000

    qwen-plus-2025-04-28

    (qwen-plus-0428)

    1,000,000

    qwen-plus-2025-01-25

    (qwen-plus-0125)

    150,000

    qwen-plus-2025-01-12

    (qwen-plus-0112)

    qwen-plus-2024-12-20

    (qwen-plus-1220)

    qwen-flash

    30,000

    10,000,000

    qwen-flash-2025-07-28

    60

    1,000,000

    qwq-plus

    600

    1,000,000

    qwq-plus-latest

    qwq-plus-2025-03-05

    60

    100,000

    qwen-turbo

    1,200

    5,000,000

    qwen-turbo-latest

    qwen-turbo-2025-04-28

    (qwen-turbo-0428)

    60

    1,000,000

    qwen-turbo-2025-02-11

    (qwen-turbo-0211)

    5,000,000

    qwen-turbo-2024-11-01

    (qwen-turbo-1101)

    qwen-long-latest

    1,200

    60,000

    qwen-long-2025-01-25

    (qwen-long-0125)

    3

    7,500

    Qwen-VL (visual understanding / image-to-text)

    Global

    In the Global deployment mode, the endpoint and data storage are located in the US (Virginia) region. Inference computing resources are dynamically scheduled worldwide.

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-vl-plus

    1,200

    1,000,000

    qwen3-vl-plus-2025-09-23

    60

    100,000

    qwen3-vl-flash

    1,200

    1,000,000

    qwen3-vl-flash-2025-10-15

    60

    100,000

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-vl-plus

    1,200

    1,000,000

    qwen3-vl-plus-2025-12-19

    60

    100,000

    qwen3-vl-plus-2025-09-23

    120

    1,000,000

    qwen3-vl-flash

    1,200

    1,000,000

    qwen3-vl-flash-2026-01-22

    60

    100,000

    qwen3-vl-flash-2025-10-15

    120

    1,000,000

    qwen-vl-max

    1,200

    1,000,000

    qwen-vl-max-latest

    qwen-vl-max-2025-08-13

    (qwen-vl-max-0813)

    60

    100,000

    qwen-vl-max-2025-04-08

    (qwen-vl-max-0408)

    1,200

    1,000,000

    qwen-vl-plus

    qwen-vl-plus-latest

    qwen-vl-plus-2025-08-15

    (qwen-vl-plus-0815)

    120

    1,000,000

    qwen-vl-plus-2025-05-07

    (qwen-vl-plus-0507)

    qwen-vl-plus-2025-01-25

    (qwen-vl-plus-0125)

    1,200

    qvq-max

    60

    100,000

    qvq-max-latest

    qvq-max-2025-03-25

    (qvq-max-0325)

    US

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-vl-flash-us

    1,200

    1,000,000

    qwen3-vl-flash-2025-10-15-us

    120

    1,000,000

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-vl-plus

    3,000

    5,000,000

    qwen3-vl-plus-2025-12-19

    60

    100,000

    qwen3-vl-plus-2025-09-23

    qwen3-vl-flash

    3,000

    5,000,000

    qwen3-vl-flash-2026-01-22

    60

    100,000

    qwen3-vl-flash-2025-10-15

    qwen-vl-max

    1,200

    1,000,000

    qwen-vl-max-latest

    qwen-vl-max-2025-08-13

    (qwen-vl-max-0813)

    60

    100,000

    qwen-vl-max-2025-04-08

    (qwen-vl-max-0408)

    qwen-vl-max-2025-04-02

    (qwen-vl-max-0402)

    qwen-vl-max-2025-01-25

    (qwen-vl-max-0125)

    qwen-vl-max-2024-12-30

    (qwen-vl-max-1230)

    qwen-vl-max-2024-11-19

    (qwen-vl-max-1119)

    qwen-vl-plus

    1,200

    1,000,000

    qwen-vl-plus-latest

    qwen-vl-plus-2025-08-15

    (qwen-vl-plus-0815)

    60

    100,000

    qwen-vl-plus-2025-07-10

    (qwen-vl-plus-0710)

    qwen-vl-plus-2025-05-07

    (qwen-vl-plus-0507)

    qwen-vl-plus-2025-01-25

    (qwen-vl-plus-0125)

    qwen-vl-plus-2025-01-02

    (qwen-vl-plus-0102)

    qvq-max

    qvq-max-latest

    qvq-max-2025-05-15

    (qvq-max-0515)

    qvq-max-2025-03-25

    (qvq-max-0325)

    qvq-plus

    qvq-plus-latest

    qvq-plus-2025-05-15

    (qvq-plus-0515)

    Qwen-Omni (omni-modality)

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-omni-flash

    60

    100,000

    qwen3-omni-flash-2025-12-01

    qwen3-omni-flash-2025-09-15

    qwen-omni-turbo

    qwen-omni-turbo-latest

    qwen-omni-turbo-2025-03-26

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-omni-flash

    60

    100,000

    qwen3-omni-flash-2025-12-01

    qwen3-omni-flash-2025-09-15

    qwen-omni-turbo

    qwen-omni-turbo-latest

    qwen-omni-turbo-2025-03-26

    (qwen-omni-turbo-0326)

    qwen-omni-turbo-2025-01-19

    (qwen-omni-turbo-0119)

    Qwen-Omni-Realtime (real-time omni-modality)

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-omni-flash-realtime

    60

    100,000

    qwen3-omni-flash-realtime-2025-12-01

    qwen3-omni-flash-realtime-2025-09-15

    qwen-omni-turbo-realtime

    10,000

    qwen-omni-turbo-realtime-latest

    qwen-omni-turbo-realtime-2025-05-08

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-omni-flash-realtime

    60

    100,000

    qwen3-omni-flash-realtime-2025-12-01

    qwen3-omni-flash-realtime-2025-09-15

    qwen-omni-turbo-realtime

    qwen-omni-turbo-realtime-latest

    qwen-omni-turbo-realtime-2025-05-08

    Qwen-OCR (text extraction)

    Global

    In the Global deployment mode, the endpoint and data storage are located in the US (Virginia) region. Inference computing resources are dynamically scheduled worldwide.

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-vl-ocr

    600

    6,000,000

    qwen-vl-ocr-2025-11-20

    1,200

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-vl-ocr

    600

    6,000,000

    qwen-vl-ocr-2025-11-20

    1,200

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-vl-ocr

    600

    6,000,000

    qwen-vl-ocr-latest

    1,200

    qwen-vl-ocr-2025-11-20

    qwen-vl-ocr-2025-04-13

    600

    qwen-vl-ocr-2024-10-28

    Qwen-Math

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-math-plus

    1,200

    1,000,000

    qwen-math-plus-latest

    qwen-math-plus-2024-09-19

    (qwen-math-plus-0919)

    60

    100,000

    qwen-math-plus-2024-08-16

    (qwen-math-plus-0816)

    10

    20,000

    qwen-math-turbo

    1200

    1,000,000

    qwen-math-turbo-latest

    qwen-math-turbo-2024-09-19

    (qwen-math-turbo-0919)

    60

    100,000

    Qwen-Coder

    Global

    In the Global deployment mode, the endpoint and data storage are located in the US (Virginia) region. Inference computing resources are dynamically scheduled worldwide.

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-coder-plus

    2,400

    2,000,000

    qwen3-coder-plus-2025-09-23

    60

    1,000,000

    qwen3-coder-plus-2025-07-22

    qwen3-coder-flash

    1,200

    qwen3-coder-flash-2025-07-28

    60

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-coder-plus

    2,400

    2,000,000

    qwen3-coder-plus-2025-09-23

    600

    1,000,000

    qwen3-coder-plus-2025-07-22

    60

    1,000,000

    qwen3-coder-flash

    600

    5,000,000

    qwen3-coder-flash-2025-07-28

    600

    5,000,000

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-coder-plus

    5,000

    5,000,000

    qwen3-coder-plus-2025-09-23

    60

    1,000,000

    qwen3-coder-plus-2025-07-22

    qwen3-coder-flash

    5,000

    5,000,000

    qwen3-coder-flash-2025-07-28

    60

    1,000,000

    qwen-coder-plus

    1,200

    qwen-coder-plus-latest

    qwen-coder-plus-2024-11-06

    (qwen-coder-plus-1106)

    120

    200,000

    qwen-coder-turbo

    1,200

    1,000,000

    qwen-coder-turbo-latest

    qwen-coder-turbo-2024-09-19

    (qwen-coder-turbo-0919)

    60

    100,000

    Qwen-MT

    Global

    In the Global deployment mode, the endpoint and data storage are located in the US (Virginia) region. Inference computing resources are dynamically scheduled worldwide.

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-mt-plus

    60

    25,000

    qwen-mt-flash

    35,000

    qwen-mt-lite

    100,000

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-mt-plus

    60

    100,000

    qwen-mt-flash

    qwen-mt-lite

    qwen-mt-turbo

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-mt-plus

    60

    25,000

    qwen-mt-flash

    35,000

    qwen-mt-lite

    100,000

    qwen-mt-turbo

    35,000

    Qwen-Doc

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-doc-turbo

    600

    3,000,000

    Qwen-Deep-Research

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-deep-research

    120

    1,200,000

    Text generation - Qwen - Open source

    Open-source Qwen language models

    Global

    In the Global deployment mode, the endpoint and data storage are located in the US (Virginia) region. Inference computing resources are dynamically scheduled worldwide.

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-next-80b-a3b-thinking

    600

    1,000,000

    qwen3-next-80b-a3b-instruct

    qwen3-235b-a22b-thinking-2507

    qwen3-235b-a22b-instruct-2507

    qwen3-30b-a3b-thinking-2507

    qwen3-30b-a3b-instruct-2507

    qwen3-235b-a22b

    qwen3-30b-a3b

    qwen3-32b

    qwen3-14b

    qwen3-8b

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-next-80b-a3b-thinking

    600

    1,000,000

    qwen3-next-80b-a3b-instruct

    qwen3-235b-a22b-thinking-2507

    qwen3-235b-a22b-instruct-2507

    qwen3-30b-a3b-thinking-2507

    5,000,000

    qwen3-30b-a3b-instruct-2507

    qwen3-235b-a22b

    1,000,000

    qwen3-32b

    qwen3-30b-a3b

    qwen3-14b

    qwen3-8b

    qwen3-4b

    qwen3-1.7b

    qwen3-0.6b

    qwen2.5-14b-instruct-1m

    1,200

    5,000,000

    qwen2.5-7b-instruct-1m

    qwen2.5-72b-instruct

    60

    150,000

    qwen2.5-32b-instruct

    qwen2.5-14b-instruct

    qwen2.5-7b-instruct

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-next-80b-a3b-thinking

    600

    1,000,000

    qwen3-next-80b-a3b-instruct

    qwen3-235b-a22b-thinking-2507

    qwen3-235b-a22b-instruct-2507

    qwen3-30b-a3b-thinking-2507

    qwen3-30b-a3b-instruct-2507

    qwen3-235b-a22b

    qwen3-30b-a3b

    qwen3-32b

    2400

    qwen3-14b

    600

    qwen3-8b

    qwen3-4b

    qwen3-1.7b

    qwen3-0.6b

    qwq-32b

    qwq-32b-preview

    1,200

    qwen2.5-72b-instruct

    qwen2.5-32b-instruct

    qwen2.5-14b-instruct

    qwen2.5-14b-instruct-1m

    5,000,000

    qwen2.5-7b-instruct

    1,000,000

    qwen2.5-7b-instruct-1m

    5,000,000

    qwen2.5-3b-instruct

    2,000,000

    qwen2.5-1.5b-instruct

    qwen2.5-0.5b-instruct

    Qwen-VL (visual understanding / image-to-text)

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-vl-32b-thinking

    60

    100,000

    qwen3-vl-32b-instruct

    qwen3-vl-30b-a3b-thinking

    qwen3-vl-30b-a3b-instruct

    qwen3-vl-8b-thinking

    qwen3-vl-8b-instruct

    qwen3-vl-235b-a22b-thinking

    qwen3-vl-235b-a22b-instruct

    qwen2.5-vl-72b-instruct

    qwen2.5-vl-32b-instruct

    qwen2.5-vl-7b-instruct

    1,200

    1,000,000

    qwen2.5-vl-3b-instruct

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-vl-32b-thinking

    600

    1,000,000

    qwen3-vl-32b-instruct

    qwen3-vl-30b-a3b-thinking

    qwen3-vl-30b-a3b-instruct

    qwen3-vl-8b-thinking

    qwen3-vl-8b-instruct

    qwen3-vl-235b-a22b-thinking

    60

    100,000

    qwen3-vl-235b-a22b-instruct

    qwen2.5-vl-72b-instruct

    qwen2.5-vl-32b-instruct

    qwen2.5-vl-7b-instruct

    1,200

    1,000,000

    qwen2.5-vl-3b-instruct

    qwen2-vl-72b-instruct

    qwen2-vl-7b-instruct

    qwen2-vl-2b-instruct

    qvq-72b-preview

    60

    100,000

    Qwen3-Omni

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen2.5-omni-7b

    60

    100,000

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen2.5-omni-7b

    60

    100,000

    Qwen3-Omni-Captioner

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-omni-30b-a3b-captioner

    60

    100,000

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-omni-30b-a3b-captioner

    60

    100,000

    Qwen-Math

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen2.5-math-72b-instruct

    1,200

    1,000,000

    qwen2.5-math-7b-instruct

    qwen2.5-math-1.5b-instruct

    Qwen-Coder

    Global

    In the Global deployment mode, the endpoint and data storage are located in the US (Virginia) region. Inference computing resources are dynamically scheduled worldwide.

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-coder-480b-a35b-instruct

    600

    1,000,000

    qwen3-coder-30b-a3b-instruct

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-coder-480b-a35b-instruct

    1,000,000

    600

    qwen3-coder-30b-a3b-instruct

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen3-coder-480b-a35b-instruct

    600

    1,000,000

    qwen3-coder-30b-a3b-instruct

    qwen2.5-coder-32b-instruct

    1,200

    qwen2.5-coder-14b-instruct

    qwen2.5-coder-7b-instruct

    qwen2.5-coder-3b-instruct

    2,000,000

    qwen2.5-coder-1.5b-instruct

    qwen2.5-coder-0.5b-instruct

    Text generation - Third-party

    DeepSeek

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    deepseek-v3.2

    15,000

    1,200,000

    deepseek-v3.2-exp

    15,000

    1,200,000

    deepseek-v3.1

    15,000

    1,200,000

    deepseek-r1-0528

    60

    100,000

    deepseek-r1

    15,000

    1,200,000

    deepseek-v3

    deepseek-r1-distill-qwen-7b

    deepseek-r1-distill-qwen-14b

    deepseek-r1-distill-qwen-32b

    deepseek-r1-distill-qwen-1.5b

    60

    100,000

    deepseek-r1-distill-llama-8b

    deepseek-r1-distill-llama-70b

    Kimi

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    kimi-k2.5

    60

    100,000

    kimi-k2-thinking

    60

    100,000

    Moonshot-Kimi-K2-Instruct

    60

    100,000

    GLM

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    glm-4.7

    1,000,000

    60

    glm-4.6

    Image generation

    Qwen-Image

    International

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

    Service

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission limit

    Concurrent tasks

    Text-to-image

    qwen-image-max

    2 per minute

    No limit for sync API

    qwen-image-max-2025-12-30

    2 per minute

    No limit for sync API

    qwen-image-plus

    2 per second

    No limit for sync API / Async API: 2

    qwen-image-plus-2026-01-09

    2 per second

    No limit for sync API

    qwen-image

    2 per second

    No limit for sync API / Async API: 2

    Image editing

    qwen-image-edit-max

    2 per minute

    No limit for sync API

    qwen-image-edit-max-2026-01-16

    2 per minute

    No limit for sync API

    qwen-image-edit-plus

    2 per second

    No limit for sync API

    qwen-image-edit-plus-2025-12-15

    2 per second

    No limit for sync API

    qwen-image-edit-plus-2025-10-30

    2 per second

    No limit for sync API

    qwen-image-edit

    2 per second

    No limit for sync API

    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.

    Service

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission limit

    Concurrent tasks

    Text-to-image

    qwen-image-max

    2 per minute

    No limit for sync API

    qwen-image-max-2025-12-30

    2 per minute

    No limit for sync API

    qwen-image-plus

    2 per second

    No limit for sync API / Async API: 2

    qwen-image-plus-2026-01-09

    2 per second

    No limit for sync API

    qwen-image

    2 per second

    No limit for sync API / Async API: 2

    Image editing

    qwen-image-edit-max

    2 per minute

    No limit for sync API

    qwen-image-edit-max-2026-01-16

    2 per minute

    No limit for sync API

    qwen-image-edit-plus

    2 per second

    No limit for sync API

    qwen-image-edit-plus-2025-12-15

    2 per second

    No limit for sync API

    qwen-image-edit-plus-2025-10-30

    2 per second

    No limit for sync API

    qwen-image-edit

    2 per second

    No limit for sync API

    Image translation

    qwen-mt-image

    1 per second

    2

    Text-to-image - Z-Image

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    z-image-turbo

    2

    No limit for sync API

    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

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    z-image-turbo

    2

    No limit for sync API

    Wan

    Global

    In the Global deployment mode, the endpoint and data storage are located in the US (Virginia) region. Inference computing resources are dynamically scheduled worldwide.

    Service

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    Text-to-image

    wan2.6-t2i

    5

    5

    Image generation

    wan2.6-image

    5

    5

    International

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

    Service

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    Text-to-image

    wan2.6-t2i

    5

    5

    wan2.5-t2i-preview

    wan2.2-t2i-flash

    2

    2

    wan2.2-t2i-plus

    wan2.1-t2i-turbo

    wan2.1-t2i-plus

    Image editing

    wan2.5-i2i-preview

    5

    5

    Image generation

    wan2.6-image

    5

    5

    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.

    Service

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    Text-to-image

    wan2.6-t2i

    1

    5

    wan2.5-t2i-preview

    5

    wanx2.0-t2i-turbo

    2

    2

    wanx2.1-t2i-turbo

    wanx2.1-t2i-plus

    wan2.2-t2i-flash

    wan2.2-t2i-plus

    General image editing

    wan2.5-i2i-preview

    5

    5

    wanx2.1-imageedit

    2

    2

    Image generation

    wan2.6-image

    5

    5

    AI试衣OutfitAnyone

    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

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    aitryon-plus

    10

    5

    aitryon-parsing-v1

    10

    No limit for sync API

    Video generation

    Wan

    Global

    In the Global deployment mode, the endpoint and data storage are located in the US (Virginia) region. Inference computing resources are dynamically scheduled worldwide.

    Service

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    Text-to-video

    wan2.6-t2v

    5

    5

    Image-to-video - first frame

    wan2.6-i2v

    Referece-to-video

    wan2.6-r2v

    International

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

    Service

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    Text-to-image

    wan2.6-t2v

    5

    5

    wan2.5-t2v-preview

    wan2.2-t2v-plus

    2

    2

    wan2.1-t2v-turbo

    wan2.1-t2v-plus

    Image-to-video - first frame

    wan2.6-i2v-flash

    5

    5

    wan2.6-i2v

    wan2.5-i2v-preview

    wan2.2-i2v-flash

    2

    2

    wan2.1-i2v-plus

    wan2.1-i2v-turbo

    wan2.2-i2v-plus

    Image-to-video - first and last frames

    wan2.2-kf2v-flash

    wan2.1-kf2v-plus

    1

    General video editing

    wan2.1-vace-plus

    2

    Referece-to-video

    wan2.6-r2v-flash

    5

    5

    wan2.6-r2v

    5

    5

    Animate image

    wan2.2-animate-move

    5

    1

    Video character swap

    wan2.2-animate-mix

    5

    1

    US

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

    Service

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    Text-to-video

    wan2.6-t2v-us

    5

    5

    Image-to-video - first frame

    wan2.6-i2v-us

    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.

    Service

    Model

    Rate limit (triggered if any value is exceeded)

    Task submission RPS limit

    Concurrent tasks

    Text-to-video

    wan2.6-t2v

    5

    5

    wan2.5-t2v-preview

    wan2.2-t2v-plus

    2

    2

    wanx2.1-t2v-turbo

    wanx2.1-t2v-plus

    Image-to-video - first frame

    wan2.6-i2v-flash

    5

    5

    wan2.6-i2v

    wan2.5-i2v-preview

    wan2.2-i2v-plus

    2

    2

    wanx2.1-i2v-turbo

    wanx2.1-i2v-plus

    Image-to-video - first and last frames

    wan2.2-kf2v-flash

    wanx2.1-kf2v-plus

    General video editing

    wanx2.1-vace-plus

    Referece-to-video

    wan2.6-r2v-flash

    5

    5

    wan2.6-r2v

    5

    5

    Digital human

    wan2.2-s2v-detect

    5

    No limit for sync API

    wan2.2-s2v

    1

    Animate image

    wan2.2-animate-move

    5

    1

    Video character swap

    wan2.2-animate-mix

    5

    1

    AnimateAnyone

    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

    Task submission RPS limit

    Concurrent tasks

    animate-anyone-detect-gen2

    5

    No limit for sync API

    animate-anyone-template-gen2

    1

    At a time, only one task is running. Other tasks are pending.

    animate-anyone-gen2

    EMO

    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

    Task submission RPS limit

    Concurrent tasks

    emo-detect-v1

    5

    No limit for sync API

    emo-v1

    1

    At a time, only one task is running. Other tasks are pending.

    LivePortrait

    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

    Task submission RPS limit

    Concurrent tasks

    liveportrait-detect

    5

    No limit for sync API

    liveportrait

    1

    At a time, only one task is running. Other tasks are pending.

    VideoRetalk

    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

    Task submission RPS limit

    Concurrent tasks

    videoretalk

    1

    1

    At a time, only one task is running. Other tasks are pending.

    Emoji

    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

    Task submission RPS limit

    Concurrent tasks

    emoji-detect-v1

    1

    No limit for sync API

    emoji-v1

    1

    At a time, only one task is running. Other tasks are pending.

    Video style transform

    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

    Task submission RPS limit

    Concurrent tasks

    video-style-transform

    2

    1

    At a time, only one task is running. Other tasks are pending.

    Speech synthesis (text-to-speech)

    Qwen speech synthesis

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Qwen3-TTS-Instruct-Flash

    Model

    RPM

    qwen3-tts-instruct-flash

    180

    qwen3-tts-instruct-flash-2026-01-26

    180

    Qwen3-TTS-VD

    Model

    RPM

    qwen3-tts-vd-2026-01-26

    180

    Qwen3-TTS-VC

    Model

    RPM

    qwen3-tts-vc-2026-01-22

    180

    Qwen3-TTS-Flash

    Model

    RPM

    qwen3-tts-flash

    180

    qwen3-tts-flash-2025-11-27

    180

    qwen3-tts-flash-2025-09-18

    10

    Mainland China

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

    Qwen3-TTS-Instruct-Flash

    Model

    RPM

    qwen3-tts-instruct-flash

    180

    qwen3-tts-instruct-flash-2026-01-26

    180

    Qwen3-TTS-VD

    Model

    RPM

    qwen3-tts-vd-2026-01-26

    180

    Qwen3-TTS-VC

    Model

    RPM

    qwen3-tts-vc-2026-01-22

    180

    Qwen3-TTS-Flash

    Model

    RPM

    qwen3-tts-flash

    180

    qwen3-tts-flash-2025-11-27

    180

    qwen3-tts-flash-2025-09-18

    10

    Qwen-TTS

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60).

    RPM

    TPM

    Includes input and output tokens

    qwen-tts

    10

    100,000

    qwen-tts-latest

    qwen-tts-2025-05-22

    qwen-tts-2025-04-10

    Qwen real-time speech synthesis

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Qwen3-TTS-Instruct-Flash-Realtime

    Model

    RPM

    qwen3-tts-instruct-flash-realtime

    180

    qwen3-tts-instruct-flash-realtime-2026-01-22

    180

    Qwen3-TTS-VD-Realtime

    Model

    RPM

    qwen3-tts-vd-realtime-2026-01-15

    180

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

    Qwen3-TTS-VC-Realtime

    Model

    RPM

    qwen3-tts-vc-realtime-2026-01-15

    180

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

    Qwen3-TTS-Flash-Realtime

    Model

    RPM

    qwen3-tts-flash-realtime

    180

    qwen3-tts-flash-realtime-2025-11-27

    180

    qwen3-tts-flash-realtime-2025-09-18

    10

    Mainland China

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

    Qwen3-TTS-Instruct-Flash-Realtime

    Model

    RPM

    qwen3-tts-instruct-flash-realtime

    180

    qwen3-tts-instruct-flash-realtime-2026-01-22

    180

    Qwen3-TTS-VD-Realtime

    Model

    RPM

    qwen3-tts-vd-realtime-2026-01-15

    180

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

    Qwen3-TTS-VC-Realtime

    Model

    RPM

    qwen3-tts-vc-realtime-2026-01-15

    180

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

    Qwen3-TTS-Flash-Realtime

    Model

    RPM

    qwen3-tts-flash-realtime

    180

    qwen3-tts-flash-realtime-2025-11-27

    180

    qwen3-tts-flash-realtime-2025-09-18

    10

    Qwen-TTS-Realtime

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60).

    RPM

    TPM

    Includes input and output tokens

    qwen-tts-realtime

    10

    100,000

    qwen-tts-realtime-latest

    qwen-tts-realtime-2025-07-15

    Qwen voice cloning

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Model

    RPM

    qwen-voice-enrollment

    180

    Mainland China

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

    Model

    RPM

    qwen-voice-enrollment

    180

    Qwen voice design

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Model

    RPM

    qwen-voice-design

    180

    Mainland China

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

    Model

    RPM

    qwen-voice-design

    180

    CosyVoice speech synthesis

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Speech synthesis

    Model

    Task submission RPS limit

    cosyvoice-v3-plus

    3

    cosyvoice-v3-flash

    Voice cloning

    Model

    Task submission RPS limit

    cosyvoice-v3-plus

    10

    The voice cloning feature has a total concurrent request limit of 10 RPS. This limit applies whether you call a single model version or multiple versions at once. This means:

    • If you only call v2, your maximum concurrent requests are 10 RPS.

    • If you call both v2 and v3, their combined requests cannot exceed 10 RPS. For example, if v2 uses 7 RPS, v3 can use a maximum of 3 RPS.

    cosyvoice-v3-flash

    Mainland China

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

    Speech synthesis

    Model

    Task submission RPS limit

    cosyvoice-v3-plus

    3

    cosyvoice-v3-flash

    cosyvoice-v2

    Voice cloning

    Model

    Task submission RPS limit

    cosyvoice-v3-plus

    10

    The voice cloning feature has a total concurrent request limit of 10 RPS. This limit applies whether you call a single model version or multiple versions at once. This means:

    • If you only call v2, your maximum concurrent requests are 10 RPS.

    • If you call both v2 and v3, their combined requests cannot exceed 10 RPS. For example, if v2 uses 7 RPS, v3 can use a maximum of 3 RPS.

    cosyvoice-v3-flash

    cosyvoice-v2

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

    Qwen3-LiveTranslate-Flash

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60).

    RPM

    TPM

    Includes input and output tokens

    qwen3-livetranslate-flash

    100

    100,000

    qwen3-livetranslate-flash-2025-12-01

    Mainland China

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60).

    RPM

    TPM

    Includes input and output tokens

    qwen3-livetranslate-flash

    100

    100,000

    qwen3-livetranslate-flash-2025-12-01

    Qwen3-LiveTranslate-Flash-Realtime

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60).

    RPM

    TPM

    Includes input and output tokens

    qwen3-livetranslate-flash-realtime

    10

    100,000

    qwen3-livetranslate-flash-realtime-2025-09-22

    Mainland China

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60).

    RPM

    TPM

    Includes input and output tokens

    qwen3-livetranslate-flash-realtime

    10

    100,000

    qwen3-livetranslate-flash-realtime-2025-09-22

    Qwen audio file recognition

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Qwen3-ASR-Flash-Filetrans

    Model

    RPM

    qwen3-asr-flash-filetrans

    100

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

    Qwen3-ASR-Flash

    Model

    RPM

    qwen3-asr-flash

    100

    qwen3-asr-flash-2025-09-08

    United States

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

    Model

    RPM

    qwen3-asr-flash-us

    100

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

    Mainland China

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

    Qwen3-ASR-Flash-Filetrans

    Model

    RPM

    qwen3-asr-flash-filetrans

    100

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

    Qwen3-ASR-Flash

    Model

    RPM

    qwen3-asr-flash

    100

    qwen3-asr-flash-2025-09-08

    Qwen real-time speech recognition

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Model

    RPS

    qwen3-asr-flash-realtime

    20

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

    Mainland China

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

    Model

    RPS

    qwen3-asr-flash-realtime

    20

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

    Paraformer speech recognition

    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

    Task submission RPS limit

    paraformer-realtime-v2

    20

    paraformer-realtime-8k-v2

    Model

    Task submission RPS limit

    Task query RPS limit

    paraformer-v2

    20

    20

    paraformer-8k-v2

    20

    Fun-ASR audio file recognition

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Model

    Task submission RPS limit

    Task query RPS limit

    fun-asr

    10

    20

    fun-asr-2025-11-07

    fun-asr-2025-08-25

    Model

    Submit job API RPM limit

    Task query RPS limit

    fun-asr-mtl

    100

    20

    fun-asr-mtl-2025-08-25

    Mainland China

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

    Model

    Task submission RPS limit

    Task query RPS limit

    fun-asr

    10

    20

    fun-asr-2025-11-07

    fun-asr-2025-08-25

    fun-asr-mtl

    fun-asr-mtl-2025-08-25

    Fun-ASR real-time speech recognition

    International

    In the international deployment mode, endpoints and data storage are in the Singapore region. Model inference compute resources are dynamically scheduled worldwide, excluding Mainland China.

    Model

    Task submission RPS limit

    fun-asr-realtime

    20

    fun-asr-realtime-2025-11-07

    Mainland China

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

    Model

    Task submission RPS limit

    fun-asr-realtime

    20

    fun-asr-realtime-2025-11-07

    fun-asr-realtime-2025-09-15

    fun-asr-flash-8k-realtime

    fun-asr-flash-8k-realtime-2026-01-28

    Text embedding

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM/Number of tasks

    Includes input and output tokens

    text-embedding-v4

    1,800

    1,000,000

    text-embedding-v3

    6,000

    24,000,000

    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

    Rate limit (triggered if any value is exceeded)

    RPS

    TPM/Number of tasks

    Includes input and output tokens

    text-embedding-v4

    30

    1,200,000

    Multimodal embedding

    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

    Rate limit

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Input tokens only

    qwen3-vl-embedding

    1,200

    600,000

    multimodal-embedding-v1

    120

    200,000

    文本排序

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    gte-rerank-v2

    5,040

    4,980,000,000

    Domain specific

    Intent recognition

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    tongyi-intent-detect-v3

    1,200

    1,000,000

    Role playing

    International

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

    Model

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-plus-character-ja

    120

    500,000

    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

    Rate limit (triggered if any value is exceeded)

    The following are per-minute limits. The service may also enforce limits based on RPS (RPM/60) and TPS (TPM/60)

    RPM

    TPM

    Includes input and output tokens

    qwen-plus-character

    120

    500,000

    已下线模型

    See Model unpublishing mechanism.

    Retired on January 30, 2026

    Category

    Model

    Rate limit (triggered if any value is exceeded)

    RPM

    TPM

    Includes input and output tokens

    Qwen-Plus

    qwen-plus-2024-11-27

    0

    0

    qwen-plus-2024-11-25

    qwen-plus-2024-09-19

    qwen-plus-2024-08-06

    Qwen-Turbo

    qwen-turbo-2024-09-19

    Qwen-VL

    qwen-vl-max-2024-10-30

    qwen-vl-max-2024-08-09

    qwen-vl-plus-2024-08-09

    Retired on August 20, 2025

    Category

    Model

    Rate limit (triggered if any value is exceeded)

    RPM

    TPM

    Includes input and output tokens

    Text generation - Qwen

    qwen2-72b-instruct

    0

    0

    qwen2-57b-a14b-instruct

    qwen2-7b-instruct

    qwen1.5-110b-chat

    qwen1.5-72b-chat

    qwen1.5-32b-chat

    qwen1.5-14b-chat

    qwen1.5-7b-chat