Rate Limits
To ensure efficient resource usage, Rate Limits are introduced to guarantee service availability and stability.
Resource Details
We implement rate limiting across the following dimensions:
- RPM: Limit on the number of requests sent per minute
- TPM: Limit on the number of input + output tokens per minute
- Concurrency: Number of requests that have been received but have not yet returned a final result
Text
| Model | u2-flash | u2 | u2-decision | kimi-k3 | glm-5.2 | u2-med |
|---|---|---|---|---|---|---|
| Limit Type | RPM / TPM | RPM / TPM | RPM / TPM | RPM / TPM | RPM / TPM | RPM / TPM |
| Limit Amount | 100 / 10,000,000 | 100 / 10,000,000 | 100 / 10,000,000 | 100 / 10,000,000 | 100 / 10,000,000 | 100 / 10,000,000 |
Speech
| Model | u2-asr | u2-tts | u2-tts-clone | u2-tts-design |
|---|---|---|---|---|
| Limit Type | Concurrency | Concurrency | Concurrency | Concurrency |
| Limit Amount | 20 | 20 | 20 | 20 |
Vision
| Model | u1-ocr | u1-ocr-med |
|---|---|---|
| Limit Type | RPM / TPM | RPM / TPM |
| Limit Amount | 100 / 100,000 | 100 / 100,000 |
Notes
1. What are rate limits?
Rate limits are restrictions placed by the API on the number of times a user can access the server or consume tokens within a specified time period. Common metrics are RPMandTPM.
2. Why do we have rate limits?
- Prevent API abuse and misuse, avoiding malicious overloads that lead to service interruptions
- Ensure fair resource allocation, preventing a single user from consuming too many resources and affecting others
- Help maintain a consistent and stable calling experience for all users
3. What happens if I hit the rate limit?
When you make too many requests or consume too many tokens/characters in a short period, the API will reject further requests until the specified time window has passed.
4. How to better use the API under rate limiting policies?
We recommend centralizing requests as much as possible: if RPM is near the limit but TPM still has capacity, you can increase throughput by batching tasks, allowing you to process more tokens under the same RPM.