# Codex usage limits explained, and what to do when you run out

AutoRouter · Oct 23, 2026 · 8 min read

As of October 2026, OpenAI estimates that ChatGPT Plus and standard Business seats get 5 to 45 local Codex messages per five-hour window on GPT-6 Astra, 15 to 160 on GPT-6.1 Sol and 350 to 3,000 on GPT-6 Luna, and says weekly limits may also apply (OpenAI). ChatGPT Pro, at $100, $200 or $500 a month, currently has no five-hour limit and no published message counts. At the limit, Codex finishes the turn it is working on; then you can wait, buy ChatGPT credits, upgrade, or sign in with an API key and pay API rates. AutoRouter, the Mac app for Codex and Claude Code, picks the model for each Codex turn so the ChatGPT plan goes to the work that needs it.

## What are the Codex usage limits on each ChatGPT plan?

Codex limits depend on your ChatGPT plan and the model you pick, and OpenAI publishes them as estimated ranges of local messages per five-hour window. Its pricing page says usage also depends on the size of the task and whether it runs locally or in the cloud. Local messages, cloud chats and ChatGPT Work all draw on one allowance.

OpenAI's Pro tiers article (updated October 2026) says Pro 200 includes more usage than Pro 100 and Pro 500 the most. New Pro 200 subscriptions get a lower allowance than before at the same price. If you had Pro 200 at any point from September 22 to 10 a.m. Pacific on September 29, 2026, you keep the previous allowance through October 29, 2026. ChatGPT Free and Go ($8 a month) plans also include Codex, with no published ranges.

## How do Codex usage limits reset?

Codex usage counts against a five-hour window, and weekly limits may also apply. Your reset times are in the usage dashboard or, in an active Codex CLI session, /status (OpenAI). OpenAI doesn't say whether the five-hour window starts with your first message or runs on a fixed clock, so take the reset time from the dashboard. On ChatGPT Pro, a model can still become unavailable until its own allowance resets; OpenAI says to use another available model meanwhile (OpenAI).

## What happens when you hit the Codex usage limit?

Codex finishes the turn it is working on, subject to fair-use limits, and the limit banner then shows your plan's options: add credits, apply an available reset, upgrade, or wait (OpenAI). By plan:

• - On ChatGPT Plus and Pro, you can buy credits and keep working without upgrading. Included usage is spent first, then credits (OpenAI).
• - On Business, the feature is blocked when your included usage runs out and the workspace has no credits, until an owner adds more (OpenAI).
• - On Enterprise and Edu, advanced features pause when the shared credit pool is empty unless owners enable overages or buy more. On agreements billed in US dollars, ask your admin about the workspace budget.

## What are your options when Codex runs out?

You can wait, buy more usage from OpenAI, move to a bigger plan, or move the session to API billing. AutoRouter publishes this blog, and its option is the last row.

With an API key, Codex bills standard API pricing to your OpenAI Platform account in place of your plan's included usage (OpenAI). To switch the Codex CLI to a key and back:

```shell
# Sign in with an API key (billed at API rates)
printenv OPENAI_API_KEY | codex login --with-api-key

# Check which sign-in is active
codex login status

# Return to your ChatGPT plan through the browser sign-in
codex logout
codex login
```

## How do you make Codex usage limits last longer?

The model matters most: on OpenAI's own estimates, a Plus window holds about 20 times as many GPT-6 Luna messages as GPT-6.1 Sol messages, and Luna's credit rates are exactly a twentieth of GPT-6.1 Sol's for input and output (OpenAI). OpenAI recommends GPT-6.1 Sol for complex coding and Luna for focused, repeatable tasks. Switch with /model in the CLI or the model control under the composer.

Here is the arithmetic for one Plus window, assuming equal-size messages. Move a quarter of your messages, such as commit messages and renames, from GPT-6.1 Sol to Luna, and the window stretches about 1.3 times (0.75 + 0.25 ÷ 20 = 0.76 of the usage). Move half and it stretches about 1.9 times (0.5 + 0.5 ÷ 20 = 0.53). OpenAI warns that similar-looking tasks can use different amounts, so treat these as estimates. Which tasks hold up on cheaper models covers where the line falls.

Fast mode uses included limits at 2.5 times the Standard rate, so GPT-6.1 Sol's 15 to 160 messages become about 6 to 64; /fast toggles it (OpenAI). OpenAI's other tips:

• - Keep prompts precise and give Codex only the relevant files.
• - Keep AGENTS.md small, and nest AGENTS.md files in large repositories to control how much context each task loads.
• - Disable MCP servers you aren't using, since each one adds context to every message.
• - Start with the default reasoning effort and raise it only when a task needs deeper planning (OpenAI).

## Can a router make the ChatGPT plan last longer in Codex?

Yes. AutoRouter, the Mac app for Codex and Claude Code, works out how hard each Codex turn is and sends it to a model that fits, so hard work gets the best model and quick work doesn't burn the plan. In Codex, OpenAI models run on your ChatGPT plan. Hard work goes to the most capable model your plans cover; everything else goes to the model that costs you least, which is often your plan or a low-cost open model. In the landing page's illustrations, a test that only fails on CI goes to GPT-6 Astra on the ChatGPT plan, and a commit message goes to a low-cost open model so the plan isn't touched. The full order is in how AutoRouter picks a model.

When the ChatGPT plan hits its limit, AutoRouter Pro ($20 a month with $20 of included usage) keeps those turns going from the included usage until the plan resets, and asks you to top up if that runs out too. Paid models bill at the provider's price plus 10%, included usage resets monthly without rolling over, and AutoRouter credits you buy never expire. Every turn shows its model, what it cost and why. AutoRouter routes only while it is the model selected, and it sometimes misjudges how hard a task is; you can rate the pick when it does. It runs on macOS 14 or later with the Codex app and CLI, and is in beta.

## Questions

### Does starting a new Codex chat reset the limit?

No. OpenAI says a new chat does not reset your usage limits, and Support can't reset ChatGPT or Codex limits either (OpenAI).

### Do ChatGPT credits expire?

Yes. Personal-plan credits are valid for 12 months from purchase (OpenAI). Business credits also last 12 months, and Enterprise and Edu terms are set in the Order Form (OpenAI). They are separate from OpenAI API credits.

### Does ChatGPT Work count against Codex limits?

Yes. OpenAI says ChatGPT Work and Codex share pricing, credits and usage limits, and on Plus and Pro, ChatGPT for Excel draws on that allowance too (OpenAI). ChatGPT's file upload, image and voice limits are separate and don't apply to Codex (OpenAI).

### Does Claude Code have the same kind of limits?

Claude Code has a similar structure with its own limits, set by Anthropic: a rolling five-hour session window on every Claude plan, plus weekly limits on paid plans (Anthropic). How Claude Code usage limits work covers them.

## Sources

• - OpenAI: Codex pricing, Codex models, Codex speed modes, Codex authentication and API pricing, checked October 22, 2026
• - OpenAI Help Center, updated October 2026: About ChatGPT Pro tiers, Using Codex with your ChatGPT plan, Using credits for flexible usage (personal plans) and Flexible pricing for Enterprise, Edu and Business
• - Anthropic: Claude pricing, checked October 22, 2026