OpenAI limits lower than you expected: read the Limits page and your usage tier

OpenAI limits lower than you expected: read the Limits page and your usage tier

The symptom

429s at traffic levels that "should" be fine, or headers showing caps far below what you assumed. No code change will fix this; the account's tier is the ceiling.

Confirm the cause

1. Check the Limits page in the dashboard. It shows your organization's actual rate limits per model. This is the source of truth, not blog posts or memory.
2. Check your usage tier. Tiers are Free, Tier 1, Tier 2, Tier 3, Tier 4, Tier 5, qualified by total paid spend (Tier 1 at $5 paid, Tier 2 at $50, Tier 3 at $100, Tier 4 at $250, Tier 5 at $1,000). Higher tiers raise both monthly usage limits and per-model rate limits.
3. Check per-model limits. Different models have different RPM and TPM caps at the same tier, and the model pages carry the numbers. A limit that binds on GPT-5 may not bind on a mini model.
4. Check project token headers. x-ratelimit-limit-project-tokens appears when a project-scoped token limit applies on top of the org limits.

The fix

• If the tier is the ceiling and the workload is legitimate, the fix is spend: tiers rise automatically as total paid spend grows.
• If one model binds, route bulk or latency-tolerant work to a model with headroom.
• Set the org's monthly usage limit deliberately so a runaway job trips a limit instead of the credit card.

Verify the fix

After the tier rises, confirm the new caps on the Limits page and in the response headers before raising job concurrency. Do not assume the upgrade; verify it.

OpenAI's rate limits guide documents the usage tiers (Free, Tier 1 through Tier 5, qualified by total paid spend: $5, $50, $100, $250, $1,000) with monthly usage limits per tier, per-model rate limits published on the model pages, and project-scoped token headers (x-ratelimit-limit-project-tokens) when a project limit applies (https://platform.openai.com/docs/guides/rate-limits).