Standard Compute
← All providers

OpenAI API Rate Limits

The OpenAI API rate-limits per organization across requests per minute/day (RPM/RPD), tokens per minute/day (TPM/TPD), images per minute, and batch-queue tokens — whichever ceiling you hit first triggers throttling. Limits grow automatically through usage tiers earned by cumulative spend.

Transcribed from the official documentation · verified 2026-07-17

How OpenAI limits work

·Limits apply at the organization/project level, not per user or per key.
·Tiers advance automatically with cumulative paid usage; each tier raises both rate limits and the monthly usage cap.
·Every model has its own thresholds; long-context variants have separate allocations, and some model families share pooled limits.
·You can hit any dimension first: a few huge requests exhaust TPM long before RPM; many tiny parallel calls do the opposite.
·Per-model, per-tier numbers are shown live in your dashboard's Limits page (they vary too often for third-party tables to stay exact).

Usage tiers

TierQualificationMonthly usage cap
FreeAllowed geography$100
Tier 1$5 paid$100
Tier 2$50 paid$500
Tier 3$100 paid$1,000
Tier 4$250 paid$5,000
Tier 5$1,000 paid$200,000

Rate limits (RPM/TPM) scale with tier and differ per model; the exact live numbers are in your account's Limits page.

When you hit the limit

·429 rate_limit_exceeded errors; check x-ratelimit-* response headers for limit, remaining, and reset values.
·Batch API overflow shows up as queue-limit rejections based on total queued input tokens per model.

Getting higher limits

·Spend your way up: tier promotion is automatic with cumulative payment history.
·Batch API moves non-interactive volume off your live limits at a discount.
·Spread load across models — limits are per model, so routing overflow to a second model keeps agents moving.

OpenAI rate limits — common questions

How do I get to OpenAI Tier 5?

Cumulative paid usage of $1,000-plus (with account age requirements) promotes you automatically. There is no application form for standard tiers; only enterprise arrangements go beyond Tier 5's limits.

Why does my agent hit OpenAI limits at 2am with no one using it?

Agents make bursts of parallel calls — retries, subagents, and background loops multiply RPM fast, and rate limits are organization-wide, so every running agent shares one pool.

Do failed requests count against rate limits?

Requests that reach the API count toward RPM even when they error, and retry storms without backoff are the classic way agents lock themselves out. Honor the reset headers with exponential backoff.

Hitting these limits from an agent? The specific errors have dedicated fixes: OpenAI “Rate limit reached for requests” (429) · 429 Too Many Requests — what it means & how to fix · Codex “You've hit your usage limit” — how to fix

And if the limits themselves are the problem: this reference is maintained by Standard Compute, a flat-rate gateway with no RPM, TPM, or daily caps — heavy use is paced, never rejected. How that works →