Claude Code weekly limit reached: keep working until reset

It is Thursday afternoon and Claude Code stops in the middle of a task:

You've hit your weekly limit · resets Mon 12:00am

This is not a per-minute 429 that clears on its own. Claude Code does not retry it, and nothing you do on your side moves the reset time.

What you can decide is how to spend the days until Monday: pay Anthropic for more Claude, or keep Claude Code running on a backend you already pay for.

Session limit, weekly limit, model limits

A Pro or Max plan runs several allowances at once, and Claude Code names the one you hit:

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

• Session limit. The five-hour window. On Claude Code 2.1.234 or later, the open session waits for the reset and then continues the task by itself.
• Weekly limit. Shared across every model and counted at the same time as the session window, so it can run out mid-session. When the reset is more than 24 hours away, Claude Code does not start that automatic wait on its own.
• Opus or Sonnet limit. Covers only requests to that model family. This is the one case where changing model helps.

Two things catch people out. The Claude apps and Claude Code draw on the same allowance, so a long claude.ai chat on Tuesday is part of why Claude Code stops on Thursday. And every subagent sends its own requests on top of the main conversation's, so a session that fans out into parallel subagents spends the week in parallel.

Read the numbers first

Run /usage in Claude Code. It shows your plan's limits and when each one resets, and on Pro or Max it breaks recent usage down by subagents, skills, plugins and MCP servers (w switches to the last seven days). That breakdown is built from this machine's history, so claude.ai chats are not in it. On claude.ai, Settings > Usage shows the same windows.

The reset time decides everything else: two hours away, take a break; three days away, read on.

What does not help

Switching model with /model. The session and weekly limits cover all models. A switch only helps when the message says Opus limit or Sonnet limit, and only to a model outside that family.

A fallback model. Claude Code's --fallback-model chain covers a model that is overloaded or unavailable. Rate-limit and billing errors never trigger it, and on a subscription every model in the chain spends the same allowance anyway.

Retrying, waiting a few minutes, or logging in again. Claude Code retries temporary 429 throttles on its own, but a spent plan limit blocks requests until the reset time, and the allowance belongs to the account, not the session.

Compacting or clearing the conversation. A smaller context makes next week last longer. It does nothing for a week that is already spent.

A second personal plan to rotate through. It adds allowance, at a risk the week is not worth. Anthropic documents separate logins for work and personal accounts (CLAUDE_CONFIG_DIR), not a second plan bought to get past the first one's cap. The consumer terms do not mention multiple accounts, but Anthropic's legal page for Claude Code says the advertised Pro and Max limits assume ordinary, individual usage, and the terms let Anthropic suspend access without notice if it believes you broke them. The stake is the accounts, not the week.

Option 1: stay on Claude and pay for the difference

The simplest way to keep the same models is to pay for usage beyond the plan.

• Usage credits. On Pro and Max, /usage-credits opens Settings > Usage on claude.ai, where you turn them on. They are prepaid, billed at standard API rates, cover both Claude Code and the Claude apps, and take a monthly spend limit. One catch: while you draw on them, the prompt cache lasts five minutes by default instead of an hour, so the first request after a pause longer than that re-reads the whole conversation.
• An API key. A Claude Console key is billed per token to that Console account, not to your plan. Set ANTHROPIC_API_KEY, approve it when Claude Code asks, and check /status. After the reset, unset ANTHROPIC_API_KEY puts you back on the subscription.

Both cost money and keep you on exactly the model you had, with nothing new to install or trust.

Option 2: keep Claude Code, change the backend

Claude Code sends its requests to whatever ANTHROPIC_BASE_URL points at, as long as it speaks Anthropic's Messages API (how that works). If you already pay for other model access -- a GLM, DeepSeek or Kimi API key, a GitHub Copilot or Kiro plan -- Claude Code can run on it until the reset.

kRouter is a local router that puts several of those backends behind one endpoint:

npm install -g @sifxprime/krouter
krouter -t

Then, in the dashboard at http://localhost:20128/dashboard:

1. Connect your providers on the Providers page.
2. On Combos, click Create Combo, name it (until-reset, say) and add models in order. When one fails with a rate limit, quota or server error, kRouter tries the next.
3. On CLI Tools, open the Claude Code card. Choose until-reset for the Claude Opus and Claude Sonnet slots and a cheap model for Claude Haiku, which then also gets Claude Code's background requests, and click Apply.

For example:

Combo "until-reset"
  1. gh/claude-sonnet-4.6     GitHub Copilot plan
  2. kr/claude-sonnet-4.5     Kiro plan
  3. glm/glm-5.3              Z.ai API key

Apply merges this into ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:20128/v1",
    "ANTHROPIC_AUTH_TOKEN": "<your-krouter-key>",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "until-reset",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "until-reset",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm/glm-5.3"
  }
}

The key comes from the Endpoint page. If you write this by hand, keep the token line: without it, Claude Code sends your claude.ai login to kRouter. With it, the login stays on your machine and your plan's limits no longer apply to these requests.

Restart Claude Code; claude -c in the same directory reopens the conversation you were in, whole transcript included. When the week resets, click Reset on the card and restart: it removes the base URL, token and model slots, which puts Claude Code back on your subscription.

Before you rely on it:

• These are different models. GLM-5.3 is not Claude, and the Claude models on Copilot and Kiro may not be the version you had. Anthropic does not support Claude Code on non-Claude models through any gateway. Claude Code on Kimi, GLM and DeepSeek covers what each offers as the agent and what breaks.
• Only three slots are mapped. The card fills the Opus, Sonnet and Haiku slots, and the default model follows the Opus one. A session on fable, or on a full model name such as claude-opus-5-5, sends that name to kRouter without a provider prefix, so it misses the combo. Run /model opus after the restart.
• Context windows differ. Claude Code assumes 200K for a model name it does not recognize, such as a combo name. The card's Max context setting writes CLAUDE_CODE_MAX_CONTEXT_TOKENS, from 200K to 1M; for less, set it by hand. Up to v0.5.163, Reset leaves that line behind, so delete it yourself. Context windows with other models has the details.
• Some features switch off. With the base URL pointing anywhere but Anthropic and kRouter's token in place, Remote Control and voice dictation are unavailable, and MCP tool search is off by default, so every MCP tool definition loads into context up front.
• Copilot and Kiro are subscriptions too. kRouter's page for each carries a risk notice: the subscription session is not licensed for proxy use, and the account may be restricted or banned. API-key providers such as GLM, DeepSeek and Kimi carry no such notice.

Option 3: let the combo fall through on its own

To make Option 2 permanent, put your Claude plan at the top of the combo:

Combo "always-on"
  1. cc/claude-sonnet-4-6     your Claude plan, connected in kRouter
  2. gh/claude-sonnet-4.6
  3. glm/glm-5.3

Requests go to your plan while it has allowance and to the next entry once Anthropic refuses them. The Quota Tracker page shows each connected Claude account's five-hour and weekly windows, plus any per-model weekly windows Anthropic reports, with reset times. Cooldowns are per model, so when only the Opus limit is hit, a Sonnet entry on the same account still answers.

Know what this involves. Connecting a Claude plan means signing in to Claude inside kRouter, so a tool other than Claude Code holds and uses your subscription token. kRouter's Claude Code provider page carries a risk notice, and Anthropic's legal page is blunter: subscription sign-in is meant for ordinary use of Claude Code and Anthropic's own apps, third-party developers may not offer Claude.ai sign-in in their own apps or store and intermediate its session tokens, and Anthropic may enforce that without notice. Option 2 keeps your login inside Claude Code.

Two more limits:

• kRouter does not see the cap coming. Routing does not use the tracker's windows to skip a Claude account in advance. kRouter finds out when a request is refused.
• It keeps checking. Anthropic marks a plan-limit refusal with its own quota headers, which kRouter does not read, so it treats the weekly refusal like any rate limit: that model on that account cools down for seconds to minutes, never more than six hours, and is then tried again. Over a three-day lockout, that means an occasional refused request before a turn falls through. If you know the account is out until Monday, switch it off with its toggle on the Quota Tracker. Turn off Empty on that page does the same for every listed connection with 5% or less left in any window, and Turn on Available brings them back.

Choosing

Common questions

When does the Claude Code weekly limit reset?

At the time printed in the message. /usage and Settings > Usage on claude.ai show it too. Go by those rather than by a fixed day of the week.

Does switching to Sonnet or Haiku get around the weekly limit?

No. The session and weekly limits are shared across all models, so /model changes nothing for them. It only helps with the separate Opus and Sonnet limits, which each cover one model family.

Do claude.ai chats count against my Claude Code weekly limit?

Yes. Anthropic counts your usage across the Claude apps and Claude Code toward the same limits, so heavy chat use earlier in the week leaves less for Claude Code.

Will Claude Code pick the task up again after the weekly reset?

Only if the session waits for it. On Claude Code 2.1.234 or later, an interactive session signed in with your plan waits on its own when the reset is less than 24 hours away. Further out, it opens its usage-limit menu instead: pick Wait here, then continue automatically, there or later from /rate-limit-options, and keep the session open.

Does kRouter raise my Claude limit?

No. Only Anthropic can add allowance to a Claude account. kRouter sends Claude Code's requests to other backends while your plan is out and, if you connect the plan, shows its windows on the Quota Tracker.

Related

• Claude Code "rate limit exceeded": every cause and the fix
• Claude Code without a subscription: 6 backends that work
• Use your GitHub Copilot plan in Claude Code and any tool
• Multiple accounts per provider: how kRouter rotates them
• Combos & fallback