Discord Download

When do Claude Code, Codex and Cursor usage limits reset?

AutoRouter8 min read

Key facts

When one of your coding agents hits a usage limit, AutoRouter, the Mac app for all your coding agents, moves on to the connected plan with the most room left before it spends any credit, and AutoRouter Pro keeps those requests going from its monthly included usage until the plan resets. Without it, you wait for the reset, pay for more usage or upgrade. Each tool resets on its own clock, as of November 19, 2026. Claude's five-hour session limit resets in full five hours after your first message, and its weekly limit resets at a fixed time each week (Anthropic). Codex counts against a five-hour window, with weekly limits that may also apply, and shows your reset time in its usage dashboard (OpenAI). Cursor's current plans refill one weekly pool, while its legacy plans reset monthly on the billing date (Cursor).

AutoRouter wrote this post. It collects each tool's limits from its maker's own pages, checked November 19, 2026, and links the detailed guide for each one.

When does each coding agent's usage limit reset?

Claude and Codex each count your usage against a five-hour limit and a longer weekly one, while Cursor counts it against one pool that refills weekly or monthly. The table lists each plan's limits, when they reset and where to see your next reset, as of November 19, 2026.

Tool and planShort limitLong limitWhen it resetsWhere to check
Claude Code on Pro or MaxFive-hour sessionWeekly, across all models, plus a separate Fable limit on some plansThe session resets in full five hours after your first message; the weekly limit at a fixed time assigned to your account (Anthropic)Settings > Usage on claude.ai, or /usage in Claude Code
Claude Code on UltraNone on verified accountsWeeklyA fixed time each week assigned to your accountSettings > Usage, or /usage
Codex on ChatGPT Plus or BusinessFive-hour window, an estimated 17 to 180 local messages on GPT-6.2 SolWeekly limits may also applyOpenAI publishes no schedule; your reset times are in the dashboard (OpenAI)The Codex usage dashboard, or /status in the Codex CLI
Codex on ChatGPT ProNoneNo published figuresShown in the dashboardThe Codex usage dashboard, or /status
Cursor on Free, Plus, Premium, Super or UltraNoneOne weekly pool for Cursor's own modelsWeekly; the dashboard shows the next date (Cursor)Usage at grok.com/dashboard
Cursor on legacy Hobby, Pro, Pro+ or UltraNoneMonthly poolsMonthly, on your billing dateThe Spending tab at cursor.com/dashboard/spending

None of the three publishes a fixed message or token count for its own limits beyond Codex's estimated ranges, so the dashboard is the only place that shows how much of a limit you have left.

How do Claude's limits work in Claude Code?

Claude Code shares one set of limits with claude.ai and Claude Desktop: every request counts against the five-hour session limit and the weekly limit at once, and whichever runs out first stops you (Claude Help Center). Anthropic publishes no message or token counts. Its pricing page sizes the plans against Pro as of November 19, 2026:

Four guides go deeper. How Claude Code usage limits work, and how to make them last covers what uses a plan up fastest and how to stretch it. What is the Claude session limit, and when does it reset? covers the five-hour window on its own. Claude 5-hour limit vs weekly limit explains which of the two stops you. How to check your Claude Code usage covers /usage and keeping usage on screen.

How do Codex limits work on each ChatGPT plan?

Codex counts local messages and cloud tasks against one allowance per five-hour window, and OpenAI publishes estimated ranges, not fixed limits (OpenAI). As of November 19, 2026, ChatGPT Plus and standard Business seats get the ranges below per five hours, depending on the model. Cloud tasks may use more of the allowance, weekly limits may also apply, and ChatGPT Pro, from $100 a month, has no five-hour limit.

Your reset times are in the Codex usage dashboard or, in a Codex CLI session, /status. At the limit, Codex finishes the request it is working on, subject to fair use. Codex usage limits explained, and what to do when you run out covers each plan and the options at the limit.

How do Cursor's usage limits work?

Cursor's current plans, sold as The X Subscription, give one weekly pool for Cursor's own models, such as Grok 4.7 and Composer 2.5, and bill third-party models from prepaid credit at API list price on Premium and above (Cursor). Legacy plans keep two monthly pools that reset on your billing date. Unused usage never rolls over on either. As of November 19, 2026, Cursor's pricing sizes the weekly pool against Plus at $8 a month: Premium at $30 gets four times as much, Super at $100 fifteen times and Ultra at $200 fifty times.

Cursor usage limits by plan, and when your usage resets covers how fast each model uses the pool and what changes when you leave a legacy plan.

What can you do when you hit a usage limit?

You can wait for the reset, pay to keep going, upgrade, or move the work to a plan that still has room. Each tool offers the first three on its own, as of November 19, 2026:

OptionClaude CodeCodexCursor
WaitFive hours from your first message for the session; your fixed weekly time for the weekly limitThe reset time in your usage dashboardThe weekly refill, or your billing date on a legacy plan
Pay to keep goingTurn on usage creditsBuy credits on Plus or Pro (OpenAI)Prepaid credit on current plans; on-demand usage on legacy plans
UpgradeMax or Ultra; new limits apply right awayA ChatGPT Pro tierPremium, Super or Ultra
Use an API keyBills every token to the key, with no hard stopRuns extra local chats at API rates (OpenAI)Not covered here

Switching tools is the other way out: a Claude limit doesn't touch your ChatGPT plan, and neither touches Cursor's pool. Doing that by hand means switching apps in the middle of a task; AutoRouter instead moves the next request to a plan with room inside the agent you are already using.

How does AutoRouter keep you working past a limit?

AutoRouter, the Mac app for all your coding agents, spreads your work across the AI subscriptions you already pay for and keeps going when one runs out. It shows up in the model picker of every coding agent you use (Claude Code, Codex, Cursor, etc.), so there is no new app to learn.

AutoRouter is free to start, and routing across your own subscriptions is free. Paid models bill at the provider's price plus 10%, which AutoRouter earns. It runs on macOS 14 or later and is in beta. How to cut what your coding agents cost covers the other ways to make a plan go further.

Questions

Which coding agents have no five-hour limit?

Claude Code on a verified Ultra account and Codex on ChatGPT Pro have no five-hour limit, as of November 19, 2026 (Anthropic, OpenAI). Both still have a weekly limit or allowance. Cursor has no five-hour limit on any plan: it counts usage against a weekly or monthly pool.

Do claude.ai chats count against Claude Code limits?

Yes. claude.ai, Claude Desktop and Claude Code count toward the same usage limit, so a long chat on the web leaves less for coding. The breakdown in /usage covers only that machine's Claude Code sessions.

Does starting a new chat reset a usage limit?

No. OpenAI says a new Codex chat doesn't reset your limits (OpenAI), and Claude's five-hour window belongs to your account, so a new chat or /clear leaves it running (Anthropic). Cursor's pool resets only on its weekly or monthly date.

Do usage limits apply when you use an API key?

No. With an API key, Claude Code bills every token to that key and has no hard stop, and it uses ANTHROPIC_API_KEY instead of your plan when the variable is set. Codex bills API-key sessions at API rates, outside your ChatGPT allowance (OpenAI).

Sources