Models

Rate limits

Rate limits are restrictions that our API imposes on the number of times a user or client can access our services within a specified period of time.

  • Rate limits can be hit across any of the options depending on what occurs first. For example, you might send 20 requests with only 100 prompt tokens to the ChatCompletions endpoint and that would fill your limit (if your request per minute limit was 20), even if you didn't send 20k tokens (if your prompt token per minute limit was 20k) within those 20 requests.
  • Rate limits are defined at the organization level, and may additionally be set at the API key level.
  • Rate limits may vary by the API and model being used.

Use the portal usage page to inspect consumption over time and the portal billing page to review your active plan, payment methods, and invoices.

Subscription tiers

The free tier is a trial with a one-time initial token quota. Requests are subject to the following per-minute limits while tokens remain. Once the quota is exhausted, enable pay-as-you-go to continue using the API.

Limit typeValue
Prompt tokens200,000/min
Completion tokens20,000/min
Cached prompt tokens500,000/min
Audio duration100 minutes/min
Requests40/min

The standard tier is a pay-as-you-go subscription and has no monthly usage limit.

Limit typeValue
Prompt tokens200,000/min
Completion tokens20,000/min
Cached prompt tokens500,000/min
Audio duration100 minutes/min
Requests100/min

For increased rate limits, please contact us.

Initial token quota

New organizations receive a one-time Initial Token Quota of 5,000,000 tokens. Promotional offers may grant a different amount. You may need to add a payment card in the portal to unlock this quota before making inference requests. Prompt tokens, completion tokens, cached prompt tokens, and audio usage are deducted from this balance. Once the balance is exhausted, inference requests on the Free tier return HTTP 402 (Payment Required). Enable pay-as-you-go on the billing page or contact us about an Enterprise plan to continue. Adding a payment method alone doesn't enable paid usage. On the Standard tier, only usage beyond the initial token quota is billed. Configured organization and API-key spending budgets remain enforced while the balance is positive. The portal shows the remaining quota as free tokens on the billing page and, while it lasts, on the usage page.

Cached prompt token conversion

Cached prompt tokens are deducted from the initial token quota by their value: a cached prompt token counts as a fraction of a prompt token equal to the ratio of their prices. For example, if cached input costs a fifth of regular input for a model, 5 cached prompt tokens are deducted as 1 token.

Audio usage conversion

Audio usage (Speech-to-Text) is deducted from the initial token quota at the following rate:

  • 1 minute of audio ≈ 2,800 tokens

Token counting & multipliers

Model multipliers weight token usage for configured monthly token limits. They don't multiply deductions from the initial token quota. Per-minute limits are configured separately for each model and endpoint.

Effective Usage = Token Count x Model Multiplier

GLM 5.3 is the baseline model with a multiplier of 1.0.

Model multipliers

ModelMultiplier
DeepSeek OCR 21.0
GLM-5.31.0
GLM-5.3-Flash1.0
gpt-oss-120b2.0
Qwen3-Embedding 4B0.5