Claude limits every paid plan in two ways: a session budget that resets every five hours, and weekly caps that reset at a fixed time each week. Both are token budgets under the hood. And claude.ai, Claude Desktop, and Claude Code all draw from the same pool — there is no separate Claude Code allowance.

That is the whole model. The rest of this post is what each piece means, why you burn through it faster than you expect, and what actually stretches it.

The 5-hour session limit

Your session-based limit resets every five hours. Settings → Usage on claude.ai shows how much of the current window you have used and when it resets. Buying extra usage does not move the clock — included limits still reset on the same five-hour cycle.

Plans scale this budget by multiplier, not by published message counts:

Plan Price Session usage
Free $0 Limited, occasional use
Pro $20/mo ($200/yr) At least 5× Free
Max 5x $100/mo 5× Pro
Max 20x $200/mo 20× Pro

Anthropic no longer publishes “hours of Sonnet per week” estimates for these plans. If you see specific hour ranges quoted, they are from 2025 docs that no longer exist — treat them as folklore.

The weekly limits

Weekly caps arrived in mid-2025 and apply on Pro, Max, Team, and seat-based Enterprise plans. There are two buckets:

  • an all-models weekly limit, and
  • a separate Opus-only weekly limit.

Both reset at a fixed day and time assigned to your account — the reset does not slide when you take a week off. Settings → Usage shows both meters and both reset times. One 2026 wrinkle: on Max plans, the Fable model runs at 50% weekly limits, so heavy Fable use drains the week roughly twice as fast.

Why you hit limits faster than you expect

The limits meter tokens, not messages or hours. Five things dominate:

Every turn resends the whole conversation. Claude is stateless. A one-line question at hour three of a session pays for every file, tool result, and answer that came before it. Cached history bills at a reduced cache-read rate (the cache lives 1 hour on subscription plans), but it still counts.

Context is expensive and invisible. Big files read into context, verbose MCP tool results, and a bloated CLAUDE.md (Anthropic suggests keeping it under ~200 lines) ride along on every single request that follows.

Model choice multiplies everything. Opus costs several times more per turn than Sonnet, and Sonnet more than Haiku. Higher thinking effort adds more on top.

Parallelism multiplies everything again. Subagents, agent teams (roughly 7× tokens in plan mode), scheduled tasks, and background work all bill to the same pool.

Cache misses after breaks. Walk away long enough for the cache to expire and the next request re-reads your whole session at full price. Claude Code’s /usage flags sessions where cache misses exceed 10% of spend.

How to check where it goes

In Claude Code, run /usage: it shows your session and weekly bars, plus attribution — tokens by skill, subagent, plugin, and per MCP server — and toggles between 24-hour and 7-day views. On claude.ai, Settings → Usage shows the same meters. /usage-credits manages extra usage.

What happens at the limit

Claude Code tells you exactly which wall you hit and when it opens:

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

Model-family caps are the soft wall — /model sonnet and you keep working. Session and all-model weekly limits are the hard wall — every model is blocked until the reset. Recent Claude Code versions can wait at the wall and resume the task automatically when the window resets (/rate-limit-options).

Getting more usage

Three options, in the order most people should consider them:

  1. Usage bundles — prepaid credits at a discount: $50 of usage for $45, $250 for $200, $1,000 for $700. Up to $2,000/month on Pro and Max. Credits apply only after plan limits run out.
  2. Usage credits — pay-as-you-go overflow billed at standard API rates, with a monthly cap you set.
  3. The API — separate billing, no session or weekly caps, per-token prices ($5/$25 per million tokens for Opus 5, $2/$10 for Sonnet 5). Predictable, but heavy agentic use usually costs more than a Max subscription.

The other lever: spend fewer tokens on the same work

Every option above buys a bigger bucket. The other lever is filling it slower. The limits meter tokens — and most tokens a coding agent sends are context the model did not need for this request.

That is the lever Halv pulls. Halv is a desktop app that compresses every request on your machine before it reaches the model — same model, same answer, roughly half the usage. It works with Claude Code on the subscription you already pay for (and with Codex, Kimi, and GLM), and a live savings meter shows exactly how many tokens each session avoided sending. The five-hour window stays five hours; you just fit about twice as much work into it.

Try it free — no card required.

FAQ

How long do Claude usage limits last before they reset? The session limit resets every five hours. Weekly limits reset at a fixed day and time assigned to your account, shown in Settings → Usage.

Do Claude.ai and Claude Code share the same usage limit? Yes. Chat, Desktop, and Claude Code all draw from one pool on your plan.

What happens when I hit my Claude limit? Claude blocks requests until the reset time it shows. Model-specific caps have an escape hatch — switch families with /model. Recent Claude Code versions can auto-wait and continue after the reset.

Can I buy more Claude usage without upgrading my plan? Yes: usage credits at standard API rates, or prepaid bundles at 10–30% off, up to $2,000/month on Pro and Max. Both kick in only after plan limits run out.

Why does one short question use so much of my limit? Every request resends the whole conversation. Short question, long context, full price (minus the cache-read discount).

Does switching to a cheaper model help with limits? Yes — Opus costs several times more per turn than Sonnet. But session and all-model weekly limits are cross-model, so switching stretches the budget rather than resetting it.

Sources

All limits above come from Anthropic’s official documentation, checked August 2026:

Limits change. If a number here disagrees with Settings → Usage, trust the meter.