ChatGPT API usage: where to read it, the tiers, and what to do at a 429

ChatGPT API usage is measured in tokens and requests, and you read it in three places: the usage object returned with every API response, the usage dashboard in your OpenAI account, and the rate limit headers on each reply. Your limits depend on your usage tier. OpenAI's rate limit guide lists three paid tiers, Build, Launch and Grow, and your organisation moves up automatically as total credit purchases reach USD 5, USD 100 and USD 500. When you go over a limit the API returns an HTTP 429 error, and the fix is to wait and retry with backoff.

This guide covers what to read, where to read it, and how to stay inside your limits. For what usage costs, see the separate guide to ChatGPT API cost. This page is about the numbers that come before the bill.

What usage is made of

Two things consume your allowance: requests and tokens. A request is one call to the API. A token is a piece of text, roughly a short word or part of a word, and both the text you send and the text you receive count. A long prompt with a short answer and a short prompt with a long answer can use the same number of tokens.

OpenAI's rate limit guide names five measures: requests per minute (RPM), requests per day (RPD), tokens per minute (TPM), tokens per day (TPD) and images per minute (IPM). Whichever limit you reach first is the one that applies. Limits are set per organisation and project, and they vary by model, so a limit on one model says nothing about another.

Where to read your usage

1. In every response

The API returns token counts in a usage field on the response object. On the Chat Completions endpoint the fields are prompt tokens, completion tokens and total tokens. The Responses endpoint names them input tokens and output tokens. Log these numbers next to each request in your own system, with a label for the feature that made the call. That one habit tells you which part of your product uses the allowance.

2. On the usage dashboard

Your OpenAI account has a usage dashboard that shows activity by day and by model, so you can see which model produces most of your tokens. Use it for the monthly view and the export, and use your own logs for the per feature view.

3. In the response headers

Replies can carry headers that report your remaining allowance in real time. The rate limit guide lists the following:

• x-ratelimit-limit-requests and x-ratelimit-limit-tokens, the limits that apply.
• x-ratelimit-remaining-requests and x-ratelimit-remaining-tokens, what is left in the current window.
• x-ratelimit-reset-requests and x-ratelimit-reset-tokens, when each one resets.
• Retry-After, the minimum number of seconds to wait before retrying a temporary rate limit error, when present.

To see the limits for each model at your tier, the guide points you to the Limits section under Organization settings in your account.

Usage tiers and how you move up

OpenAI documents three paid usage tiers:

Tier	Total credit purchases needed
Build	USD 5
Launch	USD 100
Grow	USD 500

The guide says your tier upgrades automatically as total credit purchases reach each threshold, and that higher tiers generally give higher rate limits across models. If you are planning a launch, buy credit ahead of the date so the tier is in place before traffic arrives, and check the Limits page for the numbers on the model you will use.

What a 429 means and what to do

An HTTP 429 response means you have exceeded a temporary rate limit. A 429 can also appear with a slow down message when your request rate rose too quickly, even if you stayed within your RPM and TPM numbers, and a hard spend limit can return a 429 as well. Read the error text, because the three causes need different fixes.

For a temporary rate limit, OpenAI's guide recommends these steps:

1. Use exponential backoff. Wait briefly after a failed request, then increase the delay after each further failure.
2. Respect Retry-After. Treat it as the minimum wait and add a small random delay, so many clients do not retry at the same moment.
3. Cap your retries. Limit the number of attempts and the total retry time. Failed requests still count toward your per minute limit, so resending continuously does not help.
4. Set max tokens close to the size you expect. A smaller maximum reduces the tokens reserved for each request.
5. Batch work that can wait. Use the Batch API for non urgent jobs, or combine several prompts into one request when you reach RPM but still have TPM to spare.

A simple usage plan for a small product

1. Log tokens per request with a feature name and a user or account id.
2. Total them daily and compare the sum with the dashboard once a week. A gap means calls are being made that your logs miss.
3. Set an alert for a daily total at about 80 percent of the figure you want to stay under.
4. Review the top three features by token use each month and ask whether each can use a shorter prompt, a smaller model or the Batch API.
5. Test at your expected peak before launch, and watch the remaining requests and tokens headers during the test.

Usage on ChatGPT itself versus the API

The API allowance described here is separate from using ChatGPT in an app or browser. A ChatGPT subscription does not include API credit, and API credit does not unlock a ChatGPT plan. If your question is about the free route to the API, the guide to the ChatGPT free API sets out what is free and what is billed, and OpenAPI and the ChatGPT API clears up a common mix up of terms.

Questions about ChatGPT API usage

How do I check my ChatGPT API usage?

Use the usage dashboard in your account for totals by day and model, and read the usage field in each API response for per request counts.

What is a token?

A token is a unit of text the model reads or writes. Both your input and the model's output are counted.

What happens if I hit the rate limit?

The API returns an HTTP 429 error. Wait, using the Retry-After header when it is present, then retry with exponential backoff.

How do I get a higher limit?

Your tier rises automatically as your total credit purchases reach the thresholds of USD 100 for Launch and USD 500 for Grow. Check the Limits page in your account for the figures at your current tier.

Do failed requests count toward my limit?

Yes. OpenAI's guide says failed requests still count toward your per minute limit, so retry with a delay and not in a tight loop.

What the model says about your product when someone asks

The usage numbers above describe a developer talking to a model. The reverse conversation matters to a business: a buyer asks a model which API, tool or provider to use, and the answer names a few. When your product is an API, a library or a developer tool, those answers decide who gets tried first. Ask "which API should I use for X?" in ChatGPT, Gemini, Perplexity and Claude and note who is named. AI Knows Us runs that question across the assistants and shows the sources each answer used, so you can see whether your documentation is among them.