Limits and billing

How CPU is metered

• What's measured is the CPU time of your program's process, not wall-clock time. Waiting for a free slot isn't charged.
• Each request is charged max(1, ceil(cpu_ms)) milliseconds, so the smallest charge is 1 ms.
• Failed programs (errors, timeouts, limits) are charged the CPU time they used. Requests turned away before running cost nothing.
• The app's playground and the try-it box on this site aren't metered.

Monthly allowance

Every account gets 600 CPU-seconds a month, reset at the start of each UTC month. On the free plan, once it's used up, API calls get 429 cpu_quota_exceeded until the reset. On pay as you go, usage past the allowance is billed, up to a hard cap of 10 CPU-hours a month. See pricing.

Before a program runs, its timeout is reserved against what's left of the allowance. Afterwards the reservation is settled to what it actually used.

Limits

timeout_ms sets the wall-clock limit per request, up to your plan's maximum. Hitting a limit ends the program with status timeout or limit_exceeded; see program errors.

Rate limit headers

Evaluations carry the per-key rate limit in their headers. RateLimit-Reset is in seconds.

RateLimit-Limit: 60
RateLimit-Remaining: 57
RateLimit-Reset: 41

Where you stand

GET /v1/limits returns your plan and current usage, and doesn't count against the rate limit.

{
  "plan": "free",
  "rate_limit": {"limit": 60, "remaining": 58, "used": 2, "reset_s": 41},
  "in_flight": {"limit": 4, "current": 0},
  "cpu": {
    "month": "2026-10",
    "quota_ms": 600000,
    "used_ms": 1290,
    "remaining_ms": 598710,
    "reset_s": 1036800
  },
  "max_timeout_ms": 5000,
  "request_id": "req_..."
}