Claude Code usage limits: the weekly and 5-hour caps by plan

Claude Code on a paid Claude plan runs against two caps: a session limit that resets every five hours, and a weekly limit that resets at one fixed time each week that is assigned to your account 1. The caps belong to your subscription, not to the terminal: usage in Claude Code, its IDE extensions and the Claude apps all drains one shared pool 2. Anthropic publishes no prompt counts; usage is weighted by message length, attachments, model choice and effort level, so every "X prompts per week" figure on a third-party page is an estimate, not a spec 3 1. The ladder in numbers: Pro at $20 per month, Max 5x at $100 per month, Max 20x at $200 per month, with the Max tiers buying 5 times and 20 times the Pro plan's per-session allowance 1 4.

The two clocks: one resets in five hours, one on your account's weekly day

The session cap is the short clock: Anthropic's Pro and Max plan pages both state that the session-based usage limit will reset every five hours 1 4. The weekly cap applies across all models and resets on a fixed day and time assigned to your account; it stays the same regardless of when you start using Claude or when your subscription begins, you receive the full allowance each cycle, and your next reset is visible in Settings, Usage 4.

Two mechanics matter more than most explainers admit. First, every request counts against the session and weekly allowances at the same time, so a single heavy burst, such as a large workflow fanout, can exhaust the weekly cap while the five-hour window still has room in it 5. Second, alongside the two shared caps sit per-model-family limits for Opus and Sonnet: hitting one blocks only that family, so switching families with /model keeps you working, while switching models never restores the shared session or weekly caps 5. Opus also draws meaningfully more of your quota per turn than Sonnet 6.

What each tier costs per unit of allowance

The plan pages print prices and multiples but never put the two together. Divide the monthly price by the per-session multiple: Pro is 20 ÷ 1 = 20, Max 5x is 100 ÷ 5 = 20, and Max 20x is 200 ÷ 20 = 10. Pro and Max 5x therefore cost the same $20 per Pro-equivalent unit of session allowance, and Max 20x is the only tier where the unit price falls, to $10 1. One honesty note: the 5x and 20x multiples are printed for the per-session allowance. The weekly cap also rises with the tier, since the Max plan promises higher usage limits than Pro, but no weekly multiple appears anywhere on the plan pages 4.

The ladder:

• Pro, $20 per month: the 1x baseline per-session allowance, a weekly cap, and Claude Code plus the Claude apps in one subscription 1 2.
• Max 5x, $100 per month: adds 5 times the Pro per-session allowance and priority access to new models 4.
• Max 20x, $200 per month: adds 20 times the Pro per-session allowance 4.

The weekly cap is not a rolling window

If your mental model is usage aging out over rolling seven days, the vendor's own pages say otherwise: weekly limits reset at a fixed time each week assigned to your account, and the full allowance returns at each reset 1. Claude Code's error reference shows the shape in the wild: the weekly-limit message carries a fixed reset stamp, for example "You've hit your weekly limit · resets Mon 12:00am" 5. The genuinely rolling arrangement exists one tier over: Enterprise seats draw from an organization pool that resets on a rolling window 6. Team and Enterprise metering is a different system, and Anthropic's Team and Enterprise guides own that question 6.

Every dated limit change in 2026

Anthropic retunes these caps often enough that a static explainer is expired on arrival. The dated record so far:

• May 6, 2026: Anthropic doubled Claude Code's five-hour rate limits for Pro, Max, Team and seat-based Enterprise plans, and removed the peak-hour reduction that cut Pro and Max allowances at busy times. The same announcement covered a deal to use all compute capacity at SpaceX's Colossus 1 data center, over 220,000 GPUs and more than 300 megawatts 7.
• May 13 to September 13, 2026: weekly limits ran 50% higher as an automatic promotion, in Claude Code only, across the CLI, IDE extensions, desktop and web. The five-hour caps were not affected, and the Claude apps' limits were unchanged 8.
• September 14, 2026: weekly limits settled at 25% above their pre-promotion level, permanently, for Pro, Max, Team and seat-based Enterprise plans 8 9.
• The arithmetic behind the headlines: 1.5 × 0.83 ≈ 1.25. A 17% cut from the promotional level lands at 25% above the pre-May baseline, which is why one event was covered as both an increase and a cut; Anthropic's own clarification called it "a 17% reduction in weekly limits on Claude Code" compared with the promotional level 9.

Where to watch for the next one: these cap changes landed on Anthropic's news page, help-center articles and Anthropic's X account 8 9, not in the Claude Code changelog, which tracks tool behavior release by release, most recently version 2.1.285 on September 29, 2026 10.

What hitting a cap looks like, and what is not a cap

Four messages name the four caps, each with its own reset stamp 5:

• "You've hit your session limit · resets 3:45pm"
• "You've hit your weekly limit · resets Mon 12:00am"
• "You've hit your Opus limit · resets 3:45pm"
• "You've hit your Sonnet limit · resets 3:45pm"

Warnings arrive early, with a message such as "You've used 85% of your session limit", and the CLI can park itself at a cap and resume on its own, showing "Usage limit reached · continuing automatically at 3:45pm" 5. Two errors look like caps but are not: "Server is temporarily limiting requests" is a short-lived throttle unrelated to your quota, and "Request rejected (429)" points at an API-key rate tier, not your subscription 5. "Usage credits required for 1M context" is an entitlement check on the million-token window, not an exhausted quota 5. Run /usage to see every window and its reset time, and /usage-credits to buy more on Pro and Max 5.

The cheaper paths, and when each one wins

• Stay on your plan and buy past the cap: usage credits let Pro, Max 5x and Max 20x subscribers continue after included usage runs out 11.
• Accept API credits mid-session: Claude Code offers to continue on Claude Console credits billed at standard API rates, separate from plan pricing; decline the prompt to stay inside your plan, and run /status to confirm which credential is active 2 5.
• Upgrade for the clock, not the discount: Max 5x costs the same per unit of session allowance as Pro, so it buys volume; Max 20x is the only tier that halves the per-unit price, at $10 per Pro-equivalent unit 4.
• Spend less inside any tier: every turn resends the whole conversation, so /clear between tasks and /compact mid-task stretch an allowance further than any setting, and planning the hard parts on Opus while executing on Sonnet keeps the expensive model off routine work 6.

Window for window against Codex and Cursor

Codex runs the same two-clock shape on ChatGPT plans: OpenAI restored a five-hour limit on Codex and ChatGPT Work for Plus subscribers in August 2026 12, with the weekly limit remaining in place 13. Cursor does not use time windows at all: plans include two monthly usage pools, Cursor Models and Other Models, that reset with the billing cycle, and beyond the included pools it charges on-demand pay-as-you-go rather than making you wait 14. The trade is simple. Claude Code and Codex cap your monthly spend at the subscription price and make you wait, or pay, for more. Cursor, with on-demand usage enabled, never makes you wait and never stops charging.

Verified September 2026 against Anthropic's plan and limits pages, the Claude Code error reference, and dated coverage of each change.