Codex usage limits
by plan, and when they reset.

How to check your Codex usage limits

OpenAI shows your limits in two places, both for your own account. The usage dashboard at chatgpt.com/codex/settings/usage lists the current limits and when each one resets, and /status in a running Codex CLI session prints what's left without leaving the terminal.

OpenAI suggests a look every week or two, to learn your pace before a limit stops you mid-task. Every other view of Codex usage, including the workspace ones, is on the Codex usage dashboard guide.

• /status: the active model, the session's token count and context left, then your remaining limits.
• /usage: token activity by day, by week or in total, or a banked reset to redeem when you have one.
• /statusline: puts the rate limits in the CLI footer, so they stay in view while Codex works.

/status          # remaining limits for this account
/usage weekly    # token activity this week
/statusline      # pick "rate limits" to keep them in the footer

Codex usage limits for each plan

Codex has no fixed message count. Each plan gets an allowance that local messages and cloud chats draw on together, and how much one message takes depends on the model, the context it carries, reasoning and tool use. Plan by plan, from OpenAI's pricing page and Help Center:

• Free and Go: GPT-6 Luna in the desktop app, subject to rollout. OpenAI publishes no numbers for either, and neither includes Codex Cloud.
• Plus and Business Standard: a five-hour allowance, estimated per model in the table below, and weekly limits may also apply.
• Pro, at all three tiers: no five-hour limit for now. OpenAI's Pro plans article compares how much usage each tier includes.
• Business Premium seats: more usage than a Standard seat and no five-hour limit. Owners can mix both seat types and move people between them.
• Enterprise and Edu: without flexible pricing, each seat gets Plus's limits for most features; with it, there is no fixed limit and usage draws on credits.
• An API key: no plan limits at all. Codex bills API rates per token and drops cloud features such as code review on GitHub.

How many Codex messages fit in five hours?

OpenAI's estimates are ranges, not quotas. Small scripts and routine functions sit near the top of a range, while long sessions that hold a lot of context sit near the bottom.

What uses up Codex limits faster

Most of what drains an allowance is a choice you can change. Six things move it:

• The model. Astra fits the fewest messages per window and Luna the most, so routine edits on Luna leave room for the hard ones.
• Speed. Fast mode spends included usage at 2.5 times the Standard rate, and Ultrafast, on the top Pro tier and some Enterprise plans, at 8 times.
• Cloud tasks, which can take more of the allowance than the same task on your machine.
• Images. Each one generated uses included limits 3 to 5 times faster, on average, than a turn without it.
• Context that rides along with every turn: a long AGENTS.md and each enabled MCP server.
• Other agentic features. ChatGPT Work, ChatGPT for Excel and Workspace Agents draw on the same allowance where your plan has them.

When do Codex usage limits reset?

Each window resets at the time the usage dashboard shows for it: the five-hour window on its own clock, the weekly limit on its reset date. Starting a new chat resets neither, and OpenAI Support doesn't reset limits on request; it steps in only when usage looks miscounted or access hasn't come back after the reset time.

Two kinds of extra reset exist. OpenAI sometimes grants eligible users a one-time reset, applied automatically and with no promise of another. Some Plus and Pro referral promotions bank a reset instead, which you redeem from the usage summary in the profile menu or with /usage in the CLI: it refills both the five-hour and the weekly window and moves your weekly reset date.

What happens when you hit the Codex usage limit?

A turn already running when you reach the limit gets to finish, within fair use. After that, the limit notice lists what your plan allows:

Credit rates per model are on Codex pricing, and the token cost calculator prices a month of your own tokens at API rates.

• Plus and Pro: buy credits and keep working without changing plans.
• Business, Enterprise and Edu on flexible pricing: draw on workspace credits, within the spend limits the workspace sets.
• Any plan: run more local chats on an API key, billed at API rates.
• Or upgrade, redeem a banked reset, or wait for the reset time.
• Enterprise billed in US dollars per token: the workspace budget and your own user limit decide, so ask your admin.

ChatGPT and Codex will share one usage limit in December

From December, Plus, Pro and Business plans get one allowance for ChatGPT and Codex together, measured in tokens rather than messages, as OpenAI folds Work's agentic features into the main ChatGPT app. Instant chat won't draw from it, and ChatGPT Work already shares Codex's limits today.

OpenAI expects most users to wait for a reset no more often than they do now. The combined remaining usage and its reset times will show in Settings > Usage, which makes that the page to watch after the switch.

How often heavy Codex use hits the limit: our numbers

One of our engineers ran 368 Codex sessions between June 21 and September 22, 2026, 188 of them subagent threads. In their logs the weekly limit reached 100% on 7 days, all in September, the month Codex's share of our merged pull requests rose; the five-hour limit reached 100% on 2 days, both in June.

Of the 6.55 billion input tokens in those sessions, 96.7% were cached. Cached input takes a tenth or a twentieth of the credits of new input, depending on the model, so keeping the context steady is what stretches a limit furthest. A limit that runs out on 7 days in one month is the sign to move that engineer to Pro or a Premium seat.

Tracking Codex limits across a team

OpenAI's own views stop at one account. On Business, owners and admins set monthly credit limits per seat type, with per-user overrides that take precedence; by default nobody has one. On eligible Enterprise workspaces with credit billing, each member sees their monthly Work and Codex allowance and recent credit history in Settings > Usage & billing in ChatGPT Desktop.

Zest adds the per-engineer view. Each engineer installs the Zest plugin for Codex (setup), and Zest shows each Codex session's model, its tokens and the highest five-hour and weekly limit use it reached, next to that engineer's Claude Code and Cursor sessions and the pull requests they merged. That shows which seats need a bigger plan before the week runs out.

What Zest doesn't do: draw a live meter while you work, or count usage that leaves no session file on the engineer's machine, such as ChatGPT Work chats in the browser. To see the same for Claude, read Claude usage tracker.