Codex usage limits and when they reset

Codex is included in every ChatGPT plan, with usage limits that depend on the plan. Plus and Standard Business have a five-hour limit, and a weekly limit may also apply. Your exact reset times are on your usage dashboard and in /status.

The limits, by plan

OpenAI adds that weekly limits may also apply, on top of the five-hour one.

OpenAI publishes the five-hour allowance as a range of messages for each model, and says plainly that the ranges are estimates, not fixed limits. The spread is wide because a message that edits one file and a message that works through a whole repository are very different sizes. The current ranges are on OpenAI's Codex pricing page.

What shares the allowance

Local messages (the CLI and the IDE extension) and cloud chats draw from the same allowance. OpenAI also says usage is shared with its other agent features, such as ChatGPT Work.

When do Codex limits reset?

OpenAI does not publish a reset rule. It points you to the two places that show your own reset times:

• The usage dashboard at chatgpt.com/codex/settings/usage: your current limits and when each one resets.
• /status in the Codex CLI: your remaining limits, during a session.

There is a third place most people do not know about. Codex writes its limits into its own session logs on your computer: after each reply it records how much of each window is used and the time that window resets. That is how a tool can show your Codex limits without signing in to anything. More on what is in the Codex logs.

What uses the limits fastest

From OpenAI's own list:

• The model. The larger models use far more of the allowance per message than the small one.
• Fast mode. It uses included usage at 2.5 times the standard rate.
• Context. Large projects, long sessions, a large AGENTS.md, and every MCP server you have switched on all add to what is sent.
• Reasoning and tool use. Prompt length alone is not a reliable guide to what a message costs.
• Image generation, which uses the allowance several times faster.

Making the limits last

• Start a new session for a new task. The figure that matters is how much the session re-reads on every turn, and it only grows.
• Shorten AGENTS.md, and in a large repository put smaller ones in the folders they apply to.
• Switch off MCP servers you are not using.
• Use a smaller model for routine work.
• Keep prompts specific: name the files, and say what form the answer should take.

What happens when you hit a limit

• If you are in the middle of a task, Codex can finish that turn.
• Buy credits. Plus and Pro users can buy additional credits and keep working without changing plan.
• Use an API key for extra local work, charged at standard API rates.
• Switch to a smaller model, which draws on the allowance more slowly.
• Wait for the reset time shown on your dashboard.

Know before you hit it

A percentage tells you how much is gone, not whether you will reach the reset. Cost Doctor reads the limits Codex writes to its logs and shows them in your menu bar with a forecast ("on pace to finish this window at about 72%"), next to your Claude Code limits if you use both, and names the session that used the most. The free costdoctor command prints the same in a terminal.

See also how to check your Codex usage.

Plan facts and prices on this page are from OpenAI's Codex pricing page and its Codex command reference, as checked on November 10, 2026. Limits and prices change; check that page for the current ones.