Codex Usage Limits: How They Work in 2026

Codex usage is not unlimited, and it is not a fixed number of messages either. When you sign in with ChatGPT, every Codex message draws from a shared allowance that refills on a schedule, and how fast you drain it depends on the model, the size of the task, and whether it runs on your machine or in the cloud.

Here is the short version, based on OpenAI's Codex pricing page as of October 2026:

• Codex is included in ChatGPT Free, Go, Plus, Pro, Business, Edu and Enterprise plans.
• Plus and Standard Business have a rolling five-hour window. OpenAI says Pro plans currently have no five-hour limit. Weekly limits may also apply.
• Local and cloud work share one allowance, and cloud tasks may use more of it.
• When you run out, Plus and Pro can buy credits, flexible-pricing workspaces can buy workspace credits, and anyone can keep working locally with an API key billed at API rates.
• To check where you stand, open the usage dashboard or run /status in the Codex CLI.

The rest of this guide explains each piece and how to make your allowance last.

How Codex usage limits work

OpenAI frames Codex limits as an allowance, not a quota of prompts. The pricing page says the number of messages you can send "depends on the model used, size and complexity of your tasks, and whether you run them locally or in the cloud."

That has a practical consequence. Two prompts that look the same can cost very different amounts. A small script fix might use a sliver of your window. A long refactor where the agent holds a lot of context, calls tools and reasons for a while uses far more per message. OpenAI lists model choice, context, reasoning, tool use, retrieval and caching as the drivers, and says prompt length alone isn't a reliable estimate.

The five-hour window and weekly limits

The pricing page publishes a table of estimated local messages per five-hour period for Plus and Standard Business, broken down by model. Each cell is a wide range, and the lightest model's range is far higher than the heaviest model's. OpenAI is explicit that these "are not fixed message limits" and points you to your usage dashboard for current limits and reset times.

Three qualifiers sit next to that table:

• Pro plans currently have no five-hour limit. Note the word "currently".
• Weekly limits may also apply, on top of the five-hour window.
• Enterprise and Edu without flexible pricing have the same per-seat limits as Plus for most features. Enterprise and Edu users with flexible pricing have no fixed rate limits; their usage scales with credits.

We're deliberately not copying the per-model numbers here. They change as models change, and the official page is the only version worth trusting.

Local tasks vs cloud tasks

Codex runs in two places. Local work happens in the ChatGPT desktop app, the Codex CLI and the IDE extension, on your own machine. Cloud work runs in OpenAI-hosted environments, and features like automatic GitHub code review and the Slack integration live there.

According to the pricing page, local messages and cloud chats share your plan's usage allowance, and cloud tasks may use more of it than local messages. If you're burning through your window faster than expected, check how much you're sending to the cloud.

What else counts against the same allowance

Codex isn't the only thing drawing from your limit:

The speed modes are the easy one to miss. The pricing page lists Fast mode at 2.5x on included subscription usage and Ultrafast at a higher multiplier still. If you turned on Fast mode for one urgent job and left it on, that alone can explain a week that ran out early.

Which plans include Codex (and is Codex unlimited?)

Every current ChatGPT plan includes Codex, but not on the same terms. The pricing page lists prices for each plan. Check it directly, because we don't reproduce numbers that can change without notice.

So, is Codex unlimited? Not on any standard plan. The closest thing is Enterprise or Edu with flexible pricing, where there are no fixed rate limits but usage is paid for through credits. Pro removes the five-hour cap for now, but weekly limits may still apply.

Codex pricing with an API key

Signing in with an API key is a separate billing model. The Codex authentication docs describe two methods: sign in with ChatGPT for subscription access, or sign in with an API key for usage-based access. The desktop app, CLI and IDE extension support both for local work. Codex cloud requires signing in with ChatGPT.

In API-key mode you pay standard API token prices, model availability follows what your key can access, and cloud features like GitHub code review and Slack aren't available. OpenAI recommends this mode for automation in shared environments like CI. It also warns that API token prices are separate from subscription usage, so don't use them to estimate how many tasks your plan includes.

How to check your Codex usage

You have two documented ways to see where you stand.

The usage dashboard. The pricing page links to a Codex usage dashboard under your Codex settings that shows current limits and reset times. OpenAI suggests checking it every week or two to understand your pace.

Inside the CLI. The pricing page says you can run /status during an active Codex CLI session to see your remaining limits. The developer commands reference describes /status as showing session configuration and token usage, including the active model and remaining context. The same reference lists /usage, which opens a usage menu. From there you can view account token activity, or go straight to it with /usage daily, /usage weekly or /usage cumulative.

If /status and /usage show different things, that's expected. One is about the current session; the other is about your account.

What happens when you hit the Codex usage limit

OpenAI says it wants you to be able to finish work already in progress. If you reach your limit during an active turn, the agent can keep working on that turn, subject to fair-use limits. You won't get cut off halfway through an edit.

After that, your options depend on your plan:

1. Wait for the reset. Check the dashboard for when your five-hour or weekly window refills.
2. Buy credits (Plus and Pro). Plus and Pro users can buy additional credits to keep working without upgrading.
3. Buy workspace credits (Business, Edu, Enterprise with flexible pricing). The workspace pays for usage past the included limits.
4. Switch to an API key for local work. All users can run extra local chats with an API key, charged at standard API rates. Cloud tasks still need ChatGPT sign-in.
5. Upgrade. Moving up a plan raises the allowance. Compare plans on the pricing page first.

If you're weighing Codex against another agent partly because of limits, our Claude Code vs Codex comparison covers how the two differ beyond usage.

How to make Codex usage last longer

OpenAI's own advice, plus what follows from how the allowance is calculated:

Pick a lighter model for routine work. The five-hour estimates vary enormously by model. Save the heaviest model for hard problems and use a smaller one for renames, tests and boilerplate.

Scope every prompt. Give precise instructions and drop context the task doesn't need. "Fix the failing test in auth.test.ts" costs less than "look at the auth code and improve it."

Limit source material. Point Codex at the relevant files, not the whole repo.

Say what output you need. Define the format and length, and separate required changes from nice-to-haves so the agent doesn't do optional work on your allowance.

Keep AGENTS.md lean. Codex reads AGENTS.md files before every task. The AGENTS.md docs explain that Codex concatenates them from the repository root down to your current directory, up to a combined cap of 32 KiB by default. Nest smaller files in the directories where they matter instead of one giant file at the root.

Turn off MCP servers you aren't using. OpenAI notes that every MCP server adds context to your messages and uses more of your limit.

Prefer local runs for small jobs. Cloud tasks may use more of the shared allowance.

Leave Fast mode off unless you need it. It spends included usage faster than Standard.

New to Codex overall? Start with our OpenAI Codex guide, which walks through the surfaces, sign-in and approval modes.

Where Taku fits

A lot of Codex usage goes on setup: getting a repo running, wiring tools, re-explaining the same project every session. If the setup step is where you keep stalling, Taku mirrors a working AI setup into a desktop workspace and runs it, so you can start from a configuration someone already got working. Taku is in Beta, and the Mac app is available now. Plans are on the pricing page.

FAQ

What is the Codex usage limit?

There isn't one fixed number. Plus and Standard Business have a five-hour allowance that OpenAI publishes as per-model estimates, and weekly limits may also apply. Your exact limits and reset times are on your usage dashboard.

Is Codex unlimited on ChatGPT Pro?

No. OpenAI says Pro plans currently have no five-hour limit, but weekly limits may still apply. Only Enterprise and Edu with flexible pricing have no fixed rate limits, and that usage is paid for through credits.

Do local and cloud Codex tasks have separate limits?

No. They share one allowance, and cloud tasks may use more of it than local messages.

How do I check my Codex usage?

Open the usage dashboard linked from the Codex pricing page, or run /status in the Codex CLI. /usage shows daily, weekly or cumulative account token activity.

What happens when I run out of Codex usage?

A turn already in progress can finish, subject to fair use. After that, you can wait for the reset, buy credits on Plus or Pro, use workspace credits on flexible-pricing plans, or keep working locally with an API key.

Does using Codex with an API key count against my ChatGPT plan?

No. API-key usage is billed at API rates and is separate from subscription usage. It also doesn't include cloud features like GitHub code review.

How much does Codex cost?

Codex comes with every current ChatGPT plan, and the API-key option bills per token. Prices differ by plan and change over time, so check OpenAI's pricing page for current numbers.

Key points

• Codex limits are an allowance shaped by model, task size and local vs cloud, not a fixed message count.
• Plus and Standard Business have a five-hour window; Pro currently doesn't; weekly limits may apply to all.
• Local and cloud work, ChatGPT Work, image generation and speed modes all draw from the same allowance.
• Check with the usage dashboard, /status or /usage.
• When you hit the limit, buy credits, wait for the reset, or switch to an API key for local work.
• Lighter models, tight prompts, lean AGENTS.md files and fewer MCP servers stretch the allowance furthest.