Skip to content
Docs/Platform/Rate limits

Rate limits

API limits come from your tier, set by what you have paid in total. Plan chat has its own limits. Tokens are not capped, credits are the limit there.

POST/v1/chat/completions
200
tier
this minute
0 / 10
requests per account
{ "error": { "message": "Rate limit: 10 requests per minute (Trial).", "code": "rate_limited" } }

API tiers

Your tier is set by lifetime payments. Top-ups and plan payments both count. Every account starts on the Trial tier, and moves up on its own when a payment takes it past a threshold. Credit we add to your account counts the same as a top-up.

TierPaid in totalRequests / minAt once
Trial$0101
Tier 1$5.00+304
Tier 2$50.00+608
Tier 3$250.00+12016

The API is metered per token on every plan. While a plan runs, bought or given, your account is on Tier 1 or higher, and plan payments count toward the tier like top-ups. Your tier is shown in Dashboard, Billing and returned as apiTier by GET /api/me.

Per-key limits

A key can carry its own rpm limit, set in Dashboard, API keys. It can only hold that key below your tier, never above it. Use it to cap a risky key, or to stop one worker from using the whole account's share. All of an account's keys share the tier's limits: the tier is the ceiling for the account, not for each key.

Chat on a plan

Pro and Max include the chat app up to a daily allowance, which resets at 00:00 UTC. One message is about 6K tokens of conversation in and a 500-token reply out, so longer conversations use the allowance faster. Web search in plan chat counts against the allowance at its API price.

PlanMessages / dayMessages / minReplies at onceKeepsReply up to
ProAbout 25010132K tokens4K tokens
MaxAbout 600202128K tokens8K tokens

Past the allowance, chat returns 429 chat_allowance_used. You wait for the reset, or choose to continue at API prices from your balance. Plans are on the pricing page.

What counts

  • Every POST /v1/chat/completions counts as one request, streaming or not, success or failure after auth.
  • The per-minute window is a sliding 60 seconds. It is not reset on the minute.
  • A request counts as in flight from the moment it starts until its response ends. A streaming reply is in flight until the stream closes.
  • Going over the per-minute limit returns 429 rate_limited, and the message states your limit, for example "Rate limit: 30 requests per minute (Tier 1)."
  • Going over the in-flight limit holds the call for up to 30 seconds while one of your requests finishes, then returns 429 concurrency_limited with a Retry-After header. The chat app refuses at once.
tokens per minuten/a
There is no token cap. Per-request output is capped by the model max.
/v1/modelspublic
No key, no limit applied.

Handling 429

Back off with jitter and retry. See errors for the policy. For bulk jobs, run a fixed worker pool sized to your tier's concurrency instead of firing everything and retrying.