Claude Code vs Codex usage limits: how they compare
Claude Code and Codex count subscription usage the same way: a five-hour window and a weekly one, on plans from $20 a month. The differences are in the details. OpenAI publishes rough message counts per five hours for Plus, while Anthropic publishes only multiples between plans. Codex Pro currently has no five-hour limit at all. Claude’s limits are shared with Claude chat, Codex’s with ChatGPT Work. And at the limit, Claude Code waits and carries on by itself, while Codex lets the turn in progress finish.
Both companies change their limits often, OpenAI as recently as October 29, 2026. Treat this as a map of how the two work, and your own usage page as the numbers.
Side by side
| Claude Code | Codex | |
|---|---|---|
| Free plan | Not included | Free and Go ($8) for light use, with no published numbers |
| $20 plan | Claude Pro | ChatGPT Plus |
| Higher plans | Max 5x ($100) and Max 20x ($200) | Pro at $100, $200 or $500 |
| Team plans | Team: Standard ($25 a month) and Premium ($125) seats | Business ($25 a month), with Premium seats |
| Windows | Five-hour and weekly on every paid plan | Five-hour and weekly on Plus and Standard Business; only weekly on Pro |
| What’s published | Multiples: Max is 5x or 20x Pro per five-hour session | Local messages per five hours on Plus, by model, from 5–45 to 350–3,000 |
| Shared with | Claude on the web, desktop and phone | ChatGPT Work; local and cloud tasks share one allowance |
| Check usage | /usage, Settings → Usage on claude.ai, the status line | /status, the usage dashboard, /statusline |
| At the limit | Waits, then continues the task after the reset | Finishes the turn in progress, then stops |
| Pay to keep going | Usage credits at API rates, with discounted bundles | Credits on Plus and Pro, paid weekly resets |
| Free resets | Rare, after incidents (April 23, 2026) | Earned resets, and frequent global resets |
| With an API key | Billed per token, no windows | Billed per token, no windows |
What works the same way
- Two windows at once. Every request counts against both the five-hour and the weekly window, and whichever runs out first stops you. A heavy afternoon can use up the week before the five-hour window resets.
- The weekly reset is per account, not a calendar day, on both. Each shows you the time.
- No fixed message count. Both say usage depends on the model, how much context each request carries, reasoning effort, and tools. Long threads, many MCP servers and big always-loaded instruction files (
CLAUDE.md,AGENTS.md) cost on every message in both. - Speed costs extra. Claude Code’s fast mode bills usage credits directly; Codex’s Fast mode counts 2.5 times against included usage and Ultrafast 8 times.
- An API key is the way out of windows for both: per-token billing, no reset to wait for.
Where they differ
What they tell you
OpenAI’s pricing page gives estimates for Plus and Standard Business: for example, about 15–160 local messages per five hours with GPT-6.1 Sol and 350–3,000 with GPT-6 Luna (the full table). Anthropic gives no counts at all, only that Max has 5x or 20x Pro’s usage per five-hour session and Pro at least 5x Free. Neither publishes weekly numbers.
The five-hour window on top plans
Codex Pro plans currently have no five-hour limit, only a weekly one, and on October 29, 2026 OpenAI doubled the limits of all Pro tiers and Business Premium. Claude Max keeps both windows, with Fable models capped at half of the weekly limit.
What shares your limits
On Claude plans, a long chat in the Claude app leaves less for Claude Code, since it’s one pool. On ChatGPT plans, OpenAI says Codex shares its limits and credits with ChatGPT Work; it doesn’t say that ordinary ChatGPT chat counts. Codex cloud tasks may use more of the allowance than local ones.
What happens at the limit
Claude Code stops, shows when the limit resets, and by default waits in the open session and continues the task on its own after the reset; where it’s rolling out, it may first finish the step it’s on. Codex lets the turn in progress finish and then stops until the reset (Claude Code, Codex).
Resets
OpenAI resets everyone’s Codex limits from time to time, often several times a month in autumn 2026, and lets you earn or buy extra resets. Anthropic doesn’t offer resets; it reset all subscribers’ limits once, on April 23, 2026, after a bug drained usage.
Checking
In Claude Code it’s /usage; /status shows your account and version. In the Codex CLI it’s the other way round for limits: /status shows what’s left, /usage your token activity. Both can show the limits in a status line (Claude Code, Codex).
Which gives you more?
It looks like a simple question. It isn’t. Anthropic publishes no absolute numbers to compare against, the two count different things (OpenAI in “local messages”, Anthropic in usage relative to plans), which is like comparing one fuel tank given in liters with another described as “five times the small one”. The model you pick also moves the result by more than the plan does: on Codex, the lightest model fits dozens of times more messages than the largest. What you can compare is your own week: run your usual work on each for a few days and watch the usage page, or the share of the window each task takes.
Two things do hold. At $100 and $200, Codex Pro has no five-hour cap, which suits long uninterrupted sessions; Claude Max has one but spreads usage across the Claude apps. And on both, a lighter model for routine edits stretches the limits further than any plan change.
Using both
The limits are separate, so many people use one agent while the other’s window resets. That works, but it also takes away the one natural pause the limit gave you. If you hit limits on both in the same afternoon, that’s a good sign the day has been long; why agent work wears you out is worth a look.
We built Meowdoro, a focus timer around the MacBook notch, for people who use both agents. It shows their sessions side by side, Codex’s five-hour and weekly limits as a small pie that turns orange from 80%, and how full each session’s context is for both agents. It doesn’t show Claude’s plan limits, because it doesn’t read your Claude login; for those use /usage or the status line (how it shows agents). Try it and tell us what’s missing.
Questions
Which has higher limits, Claude Code or Codex?
It can’t be compared directly: Anthropic publishes only multiples between its own plans, and the model you choose matters more than the plan. At $100 and $200, Codex Pro has no five-hour limit; Claude Max does.
Does Codex have a weekly limit like Claude Code?
Yes. Both have a weekly window on top of the five-hour one, except that Codex Pro plans currently have only the weekly one.
Is Claude Code included in the free plan?
No, Claude Code comes with paid plans. Codex is available on ChatGPT Free and Go for light use.
Do Claude Code and Codex share limits?
No. They’re different companies’ subscriptions with separate limits. Claude Code shares its limits with the Claude apps, Codex with ChatGPT Work.
What should I do when I run out on both?
Wait for the earlier reset, or switch to an API key for urgent work. Or take the break: a five-hour reset is about the length of an afternoon.
Related guides
Sources: Claude pricing; the Claude Code documentation on costs, errors, commands and interactive mode; Anthropic’s help center articles on Claude Code with Pro or Max and Fable models and its April 23 postmortem; OpenAI’s Codex pricing documentation and CLI slash commands; a post on X by Thibault Sottiaux of OpenAI on October 29, 2026. Read on October 30, 2026.