Codex usage limits: what each plan gets, how to check usage and what happens at the limit

Codex usage limits depend on your ChatGPT plan and the model you use. As of October 2026, OpenAI estimates Plus and Standard Business users get 15 to 160 local messages per five-hour window with GPT-6.1 Sol and 350 to 3,000 with GPT-6 Luna, weekly limits may also apply, and Pro plans currently have no five-hour limit. When you hit a limit, you can buy credits or switch to an API key.

Everything below comes from OpenAI's Codex pricing page and related docs, read on 26 October 2026. OpenAI changes these numbers often, so treat them as a snapshot and check your own dashboard. For plan prices, see the Codex pricing guide.

How do Codex usage limits work?

Codex doesn't give you a fixed number of messages. OpenAI says the number you can send "depends on the model used, size and complexity of your tasks, and whether you run them locally or in the cloud." A small script may use a fraction of your allowance, while long tasks that hold a lot of context use much more per message.

A few rules apply to everyone signed in with ChatGPT:

• The main window is five hours. OpenAI publishes estimates per five-hour period, and notes that "Weekly limits may also apply."
• Local and cloud share one pool. Local messages and cloud chats draw from the same plan allowance, and cloud tasks may use more of it.
• ChatGPT Work shares it too. OpenAI's pricing page says ChatGPT Work "uses the same pricing, credits, and usage limits as Codex."
• Prompt length isn't a reliable guide. Model choice, context, reasoning, tool use, retrieval and caching all affect how much a turn costs.

API key usage is different: it is billed per token at API rates and doesn't touch your ChatGPT allowance.

What are the Codex limits for each plan?

OpenAI publishes estimates for Plus and Standard Business only. These are local messages per five-hour window, as of October 2026:

OpenAI is explicit that "These estimates are not fixed message limits." The ranges are wide because a short question and a long refactor count very differently.

For the other plans, the pricing page says:

Pick the model with your limit in mind. GPT-6 Luna's range is many times higher than GPT-6.1 Sol's, so lighter tasks go much further on Luna. The Codex models guide explains which model suits which work.

What uses up your limit faster?

Some settings multiply how much of your allowance a turn uses. From the pricing page and OpenAI's speed docs:

Other things that drain the pool faster:

• Image generation uses included limits about 3 to 5 times faster on average than similar turns without images, depending on quality and size.
• Cloud tasks may use more than local messages.
• Large context. Every MCP server adds context to your messages, and so does a long AGENTS.md.
• Desktop voice uses your Codex usage budget at $0.05 per minute as of October 2026, and the model handling your task is billed separately at its token rates.

Code review has its own rule: Code Review usage applies only to reviews Codex runs through GitHub, such as tagging @codex on a pull request or automatic reviews. Reviews you run locally with /review count toward your general limits. The Codex review guide covers both.

How do you check your Codex usage?

There are four places to look:

1. The usage dashboard. Go to chatgpt.com/codex/settings/usage for your current limits and reset times. OpenAI suggests checking it "every week or two to understand your pace and remaining capacity."
2. /status in the CLI. The pricing page says you can use it to see your remaining limits during an active CLI session. It also shows the active model, approval policy and token usage.
3. /usage in the CLI. OpenAI's CLI reference says it opens a usage menu where you can view token activity or redeem an available earned reset. Jump straight to a view with /usage daily, /usage weekly or /usage cumulative.
4. The desktop app and IDE extension. In both, /status shows "the chat ID, context usage, and rate limits." In the desktop app, /usage opens usage and billing settings when you sign in with ChatGPT.

To watch your limits all the time in the CLI, run /statusline and add the rate limits item to the footer.

What happens when you hit the Codex usage limit?

OpenAI says it wants you "to be able to complete work already in progress." If you reach your limit during an active turn, Codex can keep working on that turn, subject to fair use limits. New turns then wait until your window resets, unless you do one of these:

• Buy credits (Plus and Pro). You can purchase additional credits to continue working without upgrading your plan.
• Buy workspace credits (Business, Enterprise and Edu with flexible pricing). Your workspace can buy more credits to continue.
• Use an API key. Anyone can run extra local chats with an API key, charged at standard API rates. Cloud features such as GitHub code review and Slack aren't available with an API key.
• Redeem an earned reset. If your account has one, the CLI's /usage menu offers it.
• Switch to a lighter model. GPT-6 Luna's estimates are far higher, so it may cover routine work while you wait.

In a Business, Enterprise or Edu workspace, admins can also set per-user usage limits and spend controls, so a limit you hit may come from your workspace rather than your plan. OpenAI's usage limits and spend controls page says exhausted limits "can pause access to eligible features."

How do Codex credits work?

Credits are the unit OpenAI uses to pay for eligible usage on credit-based plans. Once you reach your included limits, available credits let you keep working. The purchase price and any discounts depend on your plan or agreement.

Credits are charged per million tokens, at different rates for input, cached input and output. As of October 2026, the pricing page lists these Standard-speed rates:

There is no separate cache-write charge. Fast and Ultrafast modes multiply these rates as shown above. OpenAI also warns that "Credit prices alone don't determine included subscription usage," so you can't convert the token table into a message count for your plan. The Codex pricing guide covers the plans and API key option in more detail.

How can you make your Codex limits last longer?

OpenAI's pricing FAQ lists these tips:

• Control the size of your prompts. Be precise, but cut unnecessary context.
• Limit source material. Share only the relevant files, and narrow sources or date ranges where you can.
• Match the output to the need. Define the format and length, and separate required work from optional polish.
• Reduce the size of AGENTS.md. In larger projects, nest AGENTS.md files in subfolders so each task loads only what it needs. See the Codex AGENTS.md guide.
• Limit MCP servers. Disable servers you don't need right now. The Codex MCP guide shows how.

Two more follow from the rules above: use Standard speed unless you need the speed, and pick a smaller model or a tighter task when that still gets a useful result. The Codex best practices guide covers task sizing.

Reusing prompts that already work also saves turns, since you skip the back-and-forth of rewriting them. Kipt keeps your prompts and skills as Markdown in a private GitHub repo you own, so a prompt that worked once is easy to find with Cmd+K and run again.