When Claude Code says you've hit your usage limit, first read which limit it names and the reset time printed next to it. A session or weekly limit blocks every model until that time; an Opus or Sonnet limit only blocks that model family, so /model gets you working again in seconds.
This is the "I'm blocked right now" page, ordered by what to do first. For how the limits work, see Claude usage limits explained; for prevention, Claude Code limits: how to stay under them.
Commands, messages and plan behaviour below were read in Anthropic's Claude Code documentation and Help Center on 2026-10-15.
Which limit did you just hit?
The message tells you, and it decides everything that follows. Anthropic's error reference lists four variants, each with its reset time (read 2026-10-15):
| Message | What it covers | Does /model help? |
|---|---|---|
You've hit your session limit · resets 3:45pm | Rolling 5-hour window, all models | No |
You've hit your weekly limit · resets Mon 12:00am | Weekly allowance, all models | No |
You've hit your Opus limit · resets 3:45pm | Opus requests only | Yes, switch outside Opus |
You've hit your Sonnet limit · resets 3:45pm | Sonnet requests only | Yes, switch outside Sonnet |
One detail catches people out. Usage counts against the session and the weekly allowance at the same time, so a heavy burst can exhaust the weekly allowance before the session window even resets. If the message says "weekly", the wait can be days.
When does it reset?
The reset time is in the message itself. For the full picture, run /usage: on Pro, Max, Team and Enterprise it shows your plan usage bars and when each limit resets (costs docs, read 2026-10-15).
The session limit is a rolling window anchored to when you started, not midnight. I covered the clock mechanics in when Claude's limit resets, and the weekly slot, which differs per account, in when the weekly limit resets.
What can you do right now?
You have five real options, in the order I'd try them. The first costs nothing, the next three cost money or switch billing, the last moves the work elsewhere.
1. Let Claude Code wait and continue on its own
Since v2.1.234, an interactive session signed in with a claude.ai subscription waits in the open terminal and resumes the task after the reset (interactive mode docs, read 2026-10-15). You'll see Continuing automatically at 3:45pm · esc to cancel. Keep the session open.
Three caveats from the same page:
- Permissions still apply. The resumed task can stop on a permission prompt while you're away.
- Weekly resets are excluded. Claude Code doesn't start the wait on its own when the reset is more than 24 hours away; the usage-limit menu offers Wait here, then continue automatically instead.
- It gives up after two re-hits. If the resumed task hits the limit again twice in a row, it stops;
/rate-limit-optionsrestarts it.
2. Switch model, if the limit is model-specific
For an Opus or Sonnet limit, run /model and pick a model outside that family. The error reference adds a cost note: each model has its own prompt cache, so the next request re-reads the whole conversation with no cache hits. On a long session, that first turn is expensive.
3. Turn on usage credits
Run /usage-credits. On Pro and Max it opens Settings → Usage on claude.ai, where you enable credits, prepay and set a monthly spend limit (costs docs, read 2026-10-15).
Once on, requests past your included limit switch to pay-as-you-go and keep working. The Help Center states credits are billed at standard API rates, separately from the subscription, and that your 5-hour and weekly limits keep resetting as usual. On Team and Enterprise, the same command sends a request to your admin.
4. Switch to an API key
An API key has no reset to wait for. The Help Center describes it as pay-as-you-go, billed per token, with "no hard stop" (read 2026-10-15). Set ANTHROPIC_API_KEY or sign in with /login, then run /status to confirm which credential is active.
Know what you're signing up for. Every token is now billed, and the costs docs note the prompt cache lifetime drops to five minutes by default on an API key, against an hour on a subscription. Short breaks start costing full input price.
5. Continue in another agent, or upgrade
Your code is on disk and in git; the conversation is the only thing locked. If you have another agent installed, write a five-line brief (goal, constraint, files that matter, what's done, what's next) and carry on there. That's also the brief you'll want for /clear later.
If you hit the ceiling several times a week, the fix is a bigger plan, not a workaround. Claude Pro vs Max limits works through when the upgrade pays.
What should you do with the stalled session?
Decide between /compact and /clear before you resume, because the cache won't be warm. The costs docs say /compact "reads the conversation it summarizes, so compacting a large context is itself a large request", while /clear costs nothing.
My rule after a block:
- Same task, short history: resume as is.
- Same task, long history full of exploration: write the brief yourself,
/clear, paste it. You stop re-paying for dead file reads on every turn. - Different task:
/clear. Always.
Compacting a cold session is the most expensive way to restart. What /compact costs has the warm-versus-cold arithmetic.
How do you make the next session last longer?
Make each turn lighter, because the limit is a token budget and an agent re-sends the whole conversation on every turn. A big test log read at turn three is paid again at turn thirty. The levers, briefly; the full method is in how to stay under the limits:
- Watch the meter. Claude Code can warn you at thresholds like
You've used 85% of your session limit, and a custom status line can show therate_limitsfields permanently. Don't start a two-hour refactor at 80%. - Read the
/usagebreakdown. It attributes usage to skills, subagents, plugins and MCP servers, and flags long context or cache misses above 10% of recent usage. /clearbetween tasks,/compactat a natural break while the cache is still warm, not when auto-compact fires mid-task.- Filter tool output before the model sees it. The model needs the failing assertion, not 600 lines of passing tests.
- Match the model to the job. Anthropic's Help Center suggests Sonnet for most coding and
/model opusplanto plan with Opus and execute with Sonnet.
Points 3 and 4 are the ones nobody keeps up on a deadline. That's what I built Tokenade for: it compacts command output, file reads and code search inside Claude Code automatically, on your own machine. Unlike usage credits, it costs nothing: it's free and open source (MIT/Apache-2.0).
On the THOL benchmark it cuts session cost by 39% on long sessions against no tool at all (methodology). On a subscription, that shows up as more work per window, not money back.
What goes wrong (anti-patterns)
Switching model on a session or weekly limit. Those cover every model; /model changes nothing except your cache.
Compacting right after the block. The cache has gone cold, so the summary pays full price on the whole history. A hand-written brief and /clear are free.
Turning on credits with no spend limit. "Set to unlimited" exists. Pick a number you'd be fine seeing on the bill.
Leaving ANTHROPIC_API_KEY set after the reset. Run /status once your window is back, or you keep paying per token for work your plan covers.
Upgrading after one bad day. Check /usage across a few weeks first.
FAQ
Can I bypass the Claude Code session limit by switching models?
No. The session and weekly limits are shared across all models, so /model doesn't restore access. Only the Opus and Sonnet limits are model-specific: switching to a model outside that family keeps you working, at the cost of a cold prompt cache on the next request.
Does Claude Code continue automatically after the limit resets?
Yes, from v2.1.234, in an interactive session signed in with a claude.ai subscription. It waits in the open session and prompts Claude to pick up the task. It doesn't start the wait on its own for a reset more than 24 hours away, and not at all for API keys, background sessions or -p runs.
How much do Claude usage credits cost?
Anthropic bills usage credits at standard API rates, separately from your subscription, and you prepay a balance with an optional monthly spend limit. Rates change, so check Anthropic's pricing page rather than a number copied into a blog post.
Does /compact or /clear reset my usage limit?
No. Both act on the context window, not on what you've already spent. They make future turns cheaper, which stretches what's left of the window, but they refund nothing.
Why did I hit the limit when I barely typed anything?
Because your agent did the typing. File reads, test logs, MCP tool definitions and the replayed conversation all count, and they dwarf your prompts. Claude usage limits explained walks through where the budget goes.
See also:
- Claude Code limits: how to stay under them: the prevention guide.
- When does Claude's limit reset?: the 5-hour clock, explained.
- When does the Claude weekly limit reset?: the weekly slot.
- Claude Pro vs Max limits: when upgrading pays.
- Claude Code /compact: what it costs: warm versus cold compaction.
Sources: Error reference: usage limits · Wait for a usage limit to reset · Manage costs effectively · Manage usage credits for paid Claude plans · Models, usage, and limits in Claude Code
Cut your AI agent's token bill.
Ranked #1 on the Token-Harness Optimizer Leaderboard. Zero config.
Tokenade cuts the token bill of AI agents: one install, zero config, and it trims what your agent sends to the model.