Claude Code vs Codex vs Cursor: Usage Limits and Prices

#How each one meters the work

All three sell a monthly plan, but they count usage on different clocks, and the clock decides when a long session stops.

Claude Code runs on your Claude subscription. Anthropic's help center puts it plainly: "Both Pro and Max plans have a five-hour session limit and a weekly limit", and those limits "are shared across Claude and Claude Code". A long conversation in the browser and a long coding session draw on the same allowance. Anthropic publishes no message counts, only multiples of Pro.

Codex runs on your ChatGPT plan. OpenAI's pricing page estimates local messages per five-hour period for Plus and Business, says "Pro plans currently have no five-hour limit", and adds "Weekly limits may also apply." ChatGPT Work and Codex share usage.

Cursor moved to new plans whose included usage "refreshes weekly". The pool covers Cursor's own models; "Third-party models like Claude, GPT, and Gemini aren't part of the included pool" and are billed at API list price from prepaid credit on Premium and above.

#Plans and prices at a glance

Prices in USD as each vendor publishes them, checked 21 October 2026. Team and enterprise tiers are left out where they add nothing to the comparison.

#Claude Code: a five-hour session and a weekly cap

Claude Pro is $20 a month in the US. Max 5x is $100 and Max 20x is $200, both monthly only. Team seats are $25 a member a month (Standard) or $125 (Premium), less when billed annually, with a two-member minimum.

Anthropic expresses the plans as multiples of Pro's per-session allowance: Max 5x gets five times, Max 20x twenty times, a Team Standard seat 1.25 times and a Premium seat 6.25 times, each applied per member. The session limit resets every five hours. The weekly limit "applies across all models" and resets at a fixed time assigned to your account, shown under Settings > Usage on claude.ai.

Because the limits are shared across models, switching with /model does not restore access after a session or weekly limit. At the limit, Claude Code may finish the step it is on, then offers four ways forward: wait for the reset, upgrade, turn on usage credits at API rates, or switch to a Claude Console account. The detail is in Claude Code usage: how to check it.

#Codex: messages per five hours, then credits

Codex is included in ChatGPT Free ($0), Go ($8), Plus ($20), Pro (plans at $100, $200 or $500 a month), Business ($20 a user a month billed annually, $25 monthly, two users minimum) and Enterprise and Edu. On Plus and standard Business seats, OpenAI publishes these estimates:

OpenAI's estimates as published on its Codex pricing page, read 21 October 2026. The page adds: "These estimates are not fixed message limits".

Cloud tasks "may use more of your allowance than local messages", images use included limits about three to five times faster, and the Fast speed mode uses 2.5 times the included usage (Ultrafast, on Pro $500 only, uses 8 times). When you reach a limit the current turn can finish, and Plus and Pro can buy ChatGPT credits to keep going; local chats can also run on an API key at API rates. The full walkthrough, including where the meter lives, is in Codex usage: how to check your limits.

#Cursor: one pool that refreshes weekly

Cursor's current plans are Free, Plus at $8 a month, Premium at $30, Super at $100 and Ultra at $200, or $80, $300, $1,000 and $2,000 a year. Included usage is given only as multiples: Premium is 4x Plus, Super 15x and Ultra 50x. Plus is limited to Cursor's own models (Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5) and has no Auto mode, Cloud Agents or third-party models. Cursor's pricing page notes that subscriptions are now sold through X Subscription Corp.

Past the included usage you can keep working from prepaid credit, wait for the weekly pool to reset, or upgrade. Usage is shown under Usage at grok.com/dashboard and in the editor's settings. Older Pro ($20), Pro+ ($60) and Ultra ($200) plans keep monthly pools with unlimited Tab completions, and on-demand usage on those is off until you turn it on.

#Copilot and Gemini CLI, for completeness

GitHub Copilot now measures usage in GitHub AI Credits, where one credit is $0.01. Pro ($10 a month) includes 1,500 a month, Pro+ ($39) 7,000 and Max ($100) 20,000; Business is $19 a seat with 1,900 per user and Enterprise $39 a seat with 3,900. Code completions and next edit suggestions are not billed in credits and stay unlimited on paid plans. The usage dashboard shows the allowance, with alerts at 75, 90 and 100 percent. Annual Pro and Pro+ subscribers who stayed on the older billing keep premium requests (300 and 1,500 a month).

Gemini CLI publishes requests per user per day by sign-in: 1,000 with Gemini Code Assist for individuals, 1,500 with Google AI Pro, 2,000 with AI Ultra, and 250 on a free Gemini API key (Flash model only). Per-minute limits exist but carry no published number, and /stats model shows usage. The project's README also says that unpaid-tier and Google One users were moved to Antigravity CLI on 18 June 2026.

#Choosing by how you actually work

If Claude is also where you write and research, remember that chat and Claude Code share one allowance: a heavy conversation day is a shorter coding day. Max raises the per-session allowance rather than removing the five-hour window.

If a five-hour wall is the thing you most want gone, OpenAI currently lists no five-hour limit on Codex Pro, though weekly limits may still apply.

If you want to switch between Claude, GPT and Gemini models inside one editor, Cursor's included pool is for its own models and the rest is billed at API price from prepaid credit, so budget for both.

If most of your use is autocomplete, Copilot's paid plans do not bill completions at all.

#Keeping the meter visible

Each vendor has a way to check: /usage in Claude Code (and Settings > Usage on claude.ai), the dashboard at chatgpt.com/codex/settings/usage or /status in a Codex CLI session, Cursor's Usage dashboard, and Copilot's usage settings. Claude Code can also print its five-hour and weekly percentages in its status line, set up in a few lines of configuration.

On a notched MacBook, NotchBay shows the running Claude Code agent's status and its 5-hour and weekly usage in the notch, estimated from the local Claude Code logs, and lets you approve or deny permission prompts there. It does not track Codex, Cursor or Copilot. Other ways to watch an agent, from hooks to menu bar trackers, are in how to monitor Claude Code on a Mac.

#Frequently asked questions

#What this means for the notch

Every limit on this page is a number you only see when you go looking for it: a dashboard tab, a slash command, a percentage in a status line. The moment it matters is mid-task, when a session reaches its five-hour wall with the work half done. A meter that stays in view, at the top edge of the screen where nothing else competes for it, turns that wall into something to plan around. That is the case for putting agent status in the notch, whichever of these tools you pay for.