Claude Code "You've hit your limit · resets ...": what the Pro and Max windows are, and how to keep working before the reset
Pro and Max subscriptions meter Claude Code in windows: a rolling five-hour session allowance and, on Max, a weekly cap. When one is used up, every prompt returns You've hit your limit · resets ... until the time shown. The threads on GitHub are not mostly about the limit existing; they are about the counter and the lock disagreeing, with people locked out at 71% or 84% shown, and about limits draining faster after plan changes. Whatever the cause, the reset time is the reset time. This entry is about the hours in between.
Last verified September 21, 2026 against anthropics/claude-code #19673: You've hit your limit while usage is still at 84% (100 comments), Anthropic docs: errors (429 rate_limit_error), GitHub: 596 Claude Code issues mentioning "hit your limit" since January 2026 · 4 min read
| Provider | Claude Code on a Pro or Max subscription (not an API key) |
| HTTP status | 429 underneath; Claude Code shows the plain message |
| Error type | rate_limit_error, This request would exceed your account's rate limit |
| Message | You've hit your limit · resets 4pm (your timezone) |
| How common | 596 issues in the Claude Code repository mention it since January 2026; the largest has 77 upvotes and 100 comments, and no reply from staff |
| Can you wait it out? | Yes, and only that. The window resets at the time shown: up to five hours for the session window, up to a week for the weekly cap |
- It is the plan, not the API. An API key with credit is metered separately and keeps working; a subscription window does not care about your API balance.
- Nothing resets early. Restarting Claude Code, switching models or logging out and in does not change the window.
- To keep working: run Claude Code on an open model through another Anthropic-format endpoint until the reset. Fix 3.
What it looks like
You've hit your limit · resets 4pm (Asia/Kuala_Lumpur)
API Error: 429 {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed your account's rate limit. Please try again later."},"request_id":"req_011CX..."}The second form is what is underneath: a 429 rate_limit_error from the API. Do not confuse it with the 529 overloaded_error, which is Anthropic's capacity and clears on its own, or with a spend cap on an API organization, which needs a Console change.
The windows, and what waiting costs
| Situation | What resets it | Cost of the wait |
|---|---|---|
| 5-hour session window used up | The time shown in the message | Up to five hours of not working |
| Weekly cap used up (Max plans) | The weekly reset, up to seven days away | Days |
| Counter says 84% but you are locked out | Nothing you can do; several open reports, no staff reply on the largest | Unknown |
| Same task on an open model here | Nothing to reset; prepaid per token | 1 to 6 cents per task in our Claude Code test |
If the counter says you have room and you are locked out anyway, you are in the most upvoted open thread in the repository. There is no workaround posted there and no staff response as of September 21, 2026. Plan upgrades do not reset an active window either; "Plan upgraded but limit is not reset in sessions" is another open report in the same repository.
Fix 1: an API key is metered separately
Claude Code can run on an Anthropic API key instead of the subscription (ANTHROPIC_API_KEY). API usage is prepaid per token at Anthropic's list prices and has its own rate tiers; it is not subject to the plan windows. For heavy use this costs more than the subscription, which is why the subscription exists, but it is the only way to keep using Claude itself during a locked window.
Fix 2: make the wait shorter next time
Long tool outputs, large files in context and auto-compaction all consume the window. Claude Code's /cost and /usage show where it went. Running background sub-agents on a cheaper model, and keeping planning conversations short, stretches the same allowance further.
Fix 3: run Claude Code on an open model until the reset
Claude Code speaks the Anthropic Messages API to whatever host ANTHROPIC_BASE_URL names. Ours serves open models on that API, and the subscription window has no say over it:
export ANTHROPIC_BASE_URL="https://api.inferenceapis.com"
export ANTHROPIC_AUTH_TOKEN="your-inferenceapis-key"
export ANTHROPIC_MODEL="zai-org/GLM-5.3[1m]"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="deepseek-ai/DeepSeek-V4.1-Flash[1m]"
claude
# back to Anthropic: open a new shell, or unset the four variablesWhat you get is GLM 5.3, DeepSeek V4 Pro or Kimi K2.7 Code running your project with the same tools, plans and files, billed per token: our Claude Code test task cost between 1 and 6 cents per run depending on the model, and Claude Code 2.1.278 finished the fix-the-tests task on GLM 5.3 in 6 turns and 3 seconds on September 21, 2026. When the window resets, open a new shell and you are back on Claude. The Claude Code guide has the per-project settings form and results for eight models.
Frequently asked questions
Does using another endpoint affect my Anthropic window?
No. Requests that never reach Anthropic are not counted by Anthropic. The window keeps counting down to its reset regardless.
Can I mix: Claude for the main model, an open model for background calls?
Not in one Claude Code process; the base URL applies to every request it makes. You can run one project on each by setting the variables in that project's .claude/settings.json.
Is my code or conversation stored when I switch?
Not here. Prompts, tool results and outputs are processed and discarded; only token counts are kept for billing. Details on /trust.
Where Inference APIs fits
Fix 3 is what we sell: while the plan window is locked, Claude Code keeps working on DeepSeek V4, GLM 5.3 or Kimi K2.7 through our Anthropic-format endpoint, for cents per task, and you switch back at the reset. They are not Claude. They finished our Claude Code test task end to end. US GPUs, zero retention, prepaid; a new account starts with $1 of credit.
Something changed or wrong? Tell us and we will re-verify the entry.