Usage Tiers, Rate Limits & Response Headers
Usage tiers — qualification and monthly usage limit
| Free | User must be in an allowed geography → $100 / month approved usage limit |
|---|---|
| Build | $5 in total credit purchases → $500 / month approved usage limit |
| Launch | $100 in total credit purchases → $5,000 / month approved usage limit |
| Grow | $500 in total credit purchases → $200,000 / month approved usage limit |
| How you move up | Automatically, as your organization's total credit purchases reach each threshold. The monthly usage limit is separate from any spend limit you set. |
| Changed on | 2026-10-06: Build, Launch, Grow replaced Tier 1 to Tier 5 |
Throttling and overload errors (since 2026-09-02)
| 429 slow_down | Traffic increased too quickly. It can fire even while you are inside your RPM and TPM limits — it measures ramp rate, not usage. Follow Retry-After when present, reduce the request rate, then increase it gradually. |
|---|---|
| 503 server_is_overloaded | The requested model is temporarily overloaded. Not caused by your traffic. Follow Retry-After when present, then retry; if it continues, lengthen the delay between attempts. |
| Ramp-rate rule of thumb | Once traffic reaches 1,000,000 input TPM, grow it by no more than 50% every 15 minutes |
Rate-limit response headers
| retry-after | Minimum seconds to wait before retrying a temporary 429 or a 503 overload, when present. It does not mean a billing or quota error will clear by retrying. |
|---|---|
| x-ratelimit-limit-requests | The RPM ceiling for this key/model. |
| x-ratelimit-limit-tokens | The TPM ceiling for this key/model. |
| x-ratelimit-remaining-requests | Requests left in the current window. |
| x-ratelimit-remaining-tokens | Tokens left in the current window. |
| x-ratelimit-reset-requests | When the request window resets. |
| x-ratelimit-reset-tokens | When the token window resets. |
| x-ratelimit-limit-project-tokens | Project-scoped token ceiling, present when a project limit applies. |
| x-ratelimit-remaining-project-tokens | Tokens left under the project-scoped limit. |
| x-ratelimit-reset-project-tokens | When the project-scoped token window resets. |
Three different things get called "rate limits," and mixing them up is the commonest source of confusion: usage tiers, which set your monthly ceiling and which per-model limits you get; the per-model RPM and TPM tables, which cap how fast you can send; and — since September — a ramp-rate check that can throttle you while you're inside both.
The tiers
Since October 6, 2026 there are three paid tiers — Build, Launch and Grow — in place of Tier 1 to Tier 5. An organization moves up automatically as its total credit purchases reach each threshold. Each tier raises the approved monthly usage limit along with throughput, and that limit is separate from the hard spend limits you can set on an organization or project, which return a 429 when tracked spend reaches them.
The per-model tables
Every model's own page publishes its RPM, TPM and batch-queue table, and the tables are not all the same shape. OpenAI's guide groups the current models by family — Astra, Sol and Terra on one table, Luna on a higher one — and the Pro models start far lower than their price suggests. GPT-5.5, GPT-5.4 and GPT-5.4 Pro also publish a separate table for long-context requests, and Ultrafast traffic has token limits of its own. On today's tables every step up raises both requests and tokens per minute; under the old ladder, Tier 2 to Tier 3 left requests-per-minute flat in most groups. The rate-limit tier planner computes all of this from each model's table.
The new errors
A 429 with the code slow_down means your traffic rose too fast, and it can fire even when you're within RPM and TPM. A 503 with server_is_overloaded means the model is temporarily overloaded — not your traffic. OpenAI's rule of thumb: once you reach 1,000,000 input tokens per minute, grow by no more than 50% every 15 minutes.
The headers
Responses report the request and token ceilings, what remains and when each resets, plus project-scoped token headers when a project limit applies. Retry-After, when present, is the minimum wait for a temporary 429 or 503 — it never means a billing or quota error will clear on its own.
The tables above list the tiers, the error codes and every header.
Verified 2026-10-28 against https://developers.openai.com/api/docs/guides/rate-limits.
Could not confirm: An account's actual limits can be raised individually by OpenAI, so your dashboard is the final authority and the tools here let you override every figure. Long-context limits: the rate-limits guide still says to view them in the developer console, but the GPT-5.5, GPT-5.4 and GPT-5.4 Pro model pages publish them behind a Long Context switch, and those are the tables shown here; no other model page has one. Until 2026-10-28 this site said they were visible only in your console.
Checked: https://developers.openai.com/api/docs/guides/rate-limits · https://developers.openai.com/api/docs/changelog · https://developers.openai.com/api/docs/guides/ultrafast-mode · https://developers.openai.com/api/docs/models/gpt-6.1-sol · https://developers.openai.com/api/docs/models/gpt-5.5 · https://developers.openai.com/api/docs/models/gpt-5.4-pro