Key takeaways

  • In ChatGPT, Settings > Usage shows your remaining allowance, reset times and credit balance, but no token count for ordinary chats.
  • Every OpenAI API response has a usage object with input_tokens, output_tokens, cached_tokens, reasoning_tokens and total_tokens.
  • The platform's Analytics > Usage dashboard totals API tokens by day, hour or minute, and the Tokenizer and input token endpoint count tokens before you send.

In the ChatGPT app, check your usage under Settings > Usage. It shows your remaining allowance, when it resets and your credit balance. The app doesn't show a token count for ordinary chats. If you use the OpenAI API, every response has a usage field with exact input and output tokens, and the platform's Analytics > Usage dashboard adds them up by day, hour or minute. Both answers come from OpenAI's help center and API reference, checked on November 15, 2026.

ChatGPT and the API count usage in different ways. ChatGPT plans count messages and allowances, and none of the OpenAI pages we read describe a token counter for everyday chats. The API bills per token, so it reports every one. Pick the section below that matches how you use OpenAI.

How to check ChatGPT usage in the app

  1. Open ChatGPT on the web or in the desktop app and sign in.
  2. Open Settings.
  3. Open Usage.

OpenAI's help center says this page shows your remaining allowance and reset times for ChatGPT Work and Codex, which share one pool of usage. If you hit a limit, it shows which one you hit. If you can buy credits on your account, the same page shows your credit balance and lets you buy more.

The page also lists apps you've connected with your ChatGPT plan. OpenAI's Sign in with ChatGPT docs say Settings > Usage shows how each connected app uses your plan. Under App limits you can cap an app at a percentage of your weekly usage. The cap is a ceiling, not extra usage: app usage still counts toward your plan.

If you use Codex, there is a second view at chatgpt.com/codex/settings/usage. In a Codex CLI session, type /status. The Codex pricing page names both. We cover them in more detail in how to check Codex token usage.

What each ChatGPT plan's limits look like

Plans set limits in messages and time windows, not tokens. As of November 15, 2026, OpenAI's plans page and pricing docs say:

Plan Price Chat limits Work and Codex limits
Free $0 Unlimited text chats, subject to abuse guardrails. Limited messages with uploads No numbers published
Go Paid, monthly Unlimited text chats, subject to abuse guardrails. More messages with tools No numbers published
Plus $20 a month "Expanded messages and uploads." Limits apply Estimated messages per five-hour window, varying by model. Weekly limits may also apply
Pro $100, $200 or $500 a month Unlimited, subject to abuse guardrails. Three usage tiers No five-hour limit currently. A weekly allowance still applies
Business $20 per user a month billed yearly, $25 billed monthly Not listed on the plans page Same estimates as Plus for standard seats
Enterprise and Edu Contact sales Not listed on the plans page No fixed limits with flexible pricing. Otherwise the same per-seat limits as Plus for most features

The help center adds that Free and Go have separate limits on file uploads, image generation, voice, data analysis and other tools, even though text chat is unlimited.

For Work and Codex, the pricing docs give Plus a range of local messages per five-hour window for each model. GPT-6.1 Sol gets 15 to 160, GPT-6 Luna 350 to 3,000 and GPT-6 Astra 5 to 45. These are estimates, not fixed numbers. Bigger and harder tasks use more.

How the five-hour and weekly windows work

The help center says a plan can have a five-hour window and a weekly window at the same time, and you need allowance left in both to keep going. A new five-hour window starts with your first message after the last one ended. That means you can run out for five hours while you still have weekly usage left.

Some features use your allowance faster. OpenAI's pricing docs say image generation uses included limits about 3 to 5 times faster on average. Fast mode uses 2.5 times the included usage, and Ultrafast mode uses 8 times.

What happens when you hit a limit

ChatGPT shows when the allowance resets, when that information is available. You can switch to another model or wait. Once your plan's included usage runs out, ChatGPT draws from your credit balance if you have one. Credits expire 12 months after purchase.

OpenAI support can't reset your limits. The help center says to contact support only if usage looks miscounted, or if access doesn't come back after the stated reset time. Eligible Plus and Pro users can buy an instant reset that restores both windows.

Enterprise and Edu workspaces

Members of eligible Enterprise and Edu workspaces get a Usage & billing section in ChatGPT Desktop. Its monthly meter shows your personal credit limit across Work and Codex, the credits you've used and what's left. Admins see credit balances and download usage reports under Billing, according to OpenAI's ChatGPT Work usage and cost docs. Those docs say reports can lag, so check the report's timestamp before you compare it with an invoice.

How to check token usage in the OpenAI API

The API gives you three views: one request, all requests over time, and a count before you send anything.

Read the usage field in each response

Every Responses API reply includes a usage object. This one is from OpenAI's API reference:

"usage": {
  "input_tokens": 328,
  "input_tokens_details": {
    "cached_tokens": 0,
    "cache_write_tokens": 0
  },
  "output_tokens": 52,
  "output_tokens_details": {
    "reasoning_tokens": 0
  },
  "total_tokens": 380
}

What each field means:

  • input_tokens: everything the model read, including instructions, conversation history and tool definitions.
  • cached_tokens: input read from the prompt cache, which costs less than fresh input. cache_write_tokens counts input written to the cache.
  • output_tokens: everything the model generated, including tokens you never see.
  • reasoning_tokens: the part of the output the model spent thinking before it answered. OpenAI's reasoning guide says this can be a few hundred tokens or tens of thousands, depending on the problem.
  • total_tokens: input plus output.

The Chat Completions API uses different names for the same counts: prompt_tokens, completion_tokens and total_tokens. Hosted tools such as web search and image generation report their own usage in a separate tool_usage field.

The visible answer is often shorter than output_tokens suggests. OpenAI's token counting guide says some models, including GPT-5 models, generate formatting tokens for tool calls and message structure that never show up in the text. They still count, and max_output_tokens limits them too.

See totals on the usage dashboard

  1. Sign in at platform.openai.com.
  2. Open Analytics, then Usage.
  3. Pick a date range, and use Filters to narrow by API key, model, user, API source or processing tier.
  4. Choose input and output token metrics. Cached input, cache writes and audio tokens have their own metrics.

You need to be an organization owner or have the Usage Dashboard permission. The help center says the dashboard shows data in UTC, recent requests can take a while to appear, and project analytics shows only the selected project. For money instead of tokens, open Analytics > Spend. To keep a copy, use Export CSV and choose Activity data. Exports go back one calendar year.

Pull usage with the Usage API

To put token counts in your own reports, call the Usage API with an Admin key from your organization settings. OpenAI's cookbook example uses GET https://api.openai.com/v1/organization/usage/completions. Set bucket_width to 1m, 1h or 1d, and add group_by with model or project_id. Without group_by, those fields come back as null. Costs come from a separate endpoint, /v1/organization/costs.

Check rate limits

Rate limits are a different thing from usage: they cap tokens and requests per minute. OpenAI's rate limits guide says to find yours under Settings > Organization > Limits. Each response also carries headers such as x-ratelimit-remaining-tokens and x-ratelimit-reset-tokens, so your code can slow down before it gets a 429 error.

Count tokens before you send a request

For a quick look at how text splits into tokens, paste it into OpenAI's Tokenizer. In code, OpenAI's open-source tiktoken library counts plain text. We ran it on November 15: the sentence "How do I check my ChatGPT token usage?" is 10 tokens in the o200k_base encoding.

import tiktoken
enc = tiktoken.get_encoding("o200k_base")
print(len(enc.encode("How do I check my ChatGPT token usage?")))  # 10

Local counts miss things. The token counting guide says tiktoken can't count images or files, and tool schemas add tokens that are hard to count locally. For the exact number, send the same payload to the input token endpoint, POST /v1/responses/input_tokens. It returns input_tokens, including the formatting tokens for message roles.

curl https://api.openai.com/v1/responses/input_tokens \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-6-astra", "input": "Tell me a joke."}'

Common problems

The dashboard total is lower than my logs. Recent requests can take a while to appear, and the dashboard uses UTC. Compare the same UTC period, filtered by API key and model.

Usage and Spend don't match. The help center says the two dashboards offer different filters, so a filter set in one may not apply in the other. Use Spend, or the Cost data export, for billed amounts.

ChatGPT says I hit a limit, but Settings > Usage shows usage left. Check both windows. You can run out of the five-hour window while the weekly one still has room.

My ChatGPT plan and my API bill are separate. They are. An API key is billed at API prices, not from a ChatGPT plan's included usage, as the Codex pricing docs put it.

Where to go next

If you also use Anthropic's tools, see how to check token usage in Claude. For Codex limits in the browser and the terminal, see ChatGPT how to check usage.

Frequently asked questions

Can I see how many tokens a ChatGPT chat used?

Not in the app. None of the OpenAI pages we read describe a token counter for everyday chats. Settings > Usage shows allowances and reset times instead.

Does ChatGPT Plus have a usage limit?

Yes. OpenAI's plans page says limits apply on Plus. For Work and Codex, Plus gets an estimated number of messages per five-hour window, and weekly limits may also apply.

Where is the OpenAI API usage dashboard?

On platform.openai.com under Analytics > Usage. You need to be an organization owner or have the Usage Dashboard permission, and the data is shown in UTC.

Why is output_tokens higher than the text I got back?

Output tokens include reasoning tokens and formatting tokens that never appear in the answer. OpenAI's docs say they all count toward usage and toward max_output_tokens.

Sources

  1. Understanding ChatGPT and Codex usage limitsOpenAI
  2. Create a model response (API reference)OpenAI
  3. ChatGPT plans and pricingOpenAI
  4. PricingOpenAI
  5. Sign in with ChatGPTOpenAI
  6. ChatGPT Work: usage and costOpenAI
  7. Using Credits for Flexible Usage in ChatGPT (Personal plans)OpenAI
  8. Reviewing Work and Codex usage and using Personal Analytics in ChatGPT DesktopOpenAI
  9. Reviewing API usage and costsOpenAI
  10. Counting tokensOpenAI
  11. Reasoning modelsOpenAI
  12. Rate limitsOpenAI
  13. How to use the Usage API and Cost API to monitor your OpenAI usageOpenAI
  14. Understanding and counting tokensOpenAI
  15. tiktokenOpenAI

Add ClaudeGPT Si as a preferred source on Google