# How Claude Code usage limits work, and how to make them last

AutoRouter · Oct 23, 2026 · 12 min read

As of October 2026, Claude Code on a Pro or Max plan counts every request against two limits at once: a session limit on a rolling five-hour window and a weekly limit, both shared with claude.ai and Claude Desktop. Max 5x and Max 20x, at $100 and $200 a month in October 2026, give five or 20 times Pro's usage per session, but Anthropic publishes no fixed message or token counts for any plan. The limits last longest when routine turns run on a cheaper model than the default, Opus 5.5, when the context stays small, and when overflow goes to usage credits instead of waiting out a reset. AutoRouter, the Mac app for Codex and Claude Code, makes that model choice for each turn, and every manual method below works without it.

## What are the Claude Code usage limits on Pro and Max?

Pro, Max 5x and Max 20x each have a five-hour session limit and a weekly limit, and Anthropic describes how much more each plan gives only as a multiplier on Pro's usage. Anthropic's pricing page says there is no fixed message count, because usage depends on the length and complexity of your conversations, the model you choose and the features you use. The Pro plan article gives no number either, so per-plan prompt or hour counts you see elsewhere are third-party estimates.

Anthropic has raised these limits several times in 2026 without publishing new figures:

• - May 6, 2026: Anthropic doubled Claude Code's five-hour limits on Pro, Max, Team and seat-based Enterprise plans, and ended the peak-hours reduction for Pro and Max.
• - May 13 to September 13, 2026: a promotion raised weekly Claude Code limits by 50%. Since September 14, 2026, weekly limits have been 25% higher than they were before it.
• - September 22, 2026: with Claude Opus 5.5, Anthropic said it is increasing five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans, without saying by how much or naming Claude Code, and gave subscribers a limit reset they can save and use later.

## How do the five-hour window and the weekly limit work together?

Every request counts against both limits at the same time, and whichever runs out first stops you. The session limit runs on a rolling five-hour window; the weekly limit resets at a fixed day and time assigned to your account, which Settings > Usage shows. Claude Code's error reference warns that one burst of heavy work, such as a large workflow fan-out, can use up the weekly allowance before the session window resets.

• - On Max, Fable 5 and Fable 5.1 can use up to half of the weekly limit, and they use it faster than other Claude models. On Pro, they run on usage credits from the first request (Anthropic, updated September 2, 2026).
• - Claude Code warns you before a window runs out, with messages such as You've used 85% of your session limit · resets 3:45pm (error reference). If you hit the limit mid-response, it may finish the current step under a small, capped allowance that counts toward your weekly limit. Anthropic says that allowance is still rolling out.

Run /usage to see both limits and when they reset. To keep them in view, add the rate_limits fields to a custom status line. Save this as ~/.claude/statusline.sh, run chmod +x on it, and make sure jq is installed:

```bash
#!/bin/bash
input=$(cat)
five=$(echo "$input" | jq -r '.rate_limits.five_hour.used_percentage // empty')
week=$(echo "$input" | jq -r '.rate_limits.seven_day.used_percentage // empty')
echo "5h ${five:-?}% · week ${week:-?}%"
```

Then point Claude Code at it in ~/.claude/settings.json. The rate_limits fields appear only for Pro and Max subscribers, after the session's first response.

```json
{
  "statusLine": {
    "type": "command",
    "command": "~/.claude/statusline.sh"
  }
}
```

## What uses up Claude Code limits fastest?

The model, the size of the context and work that runs while you aren't watching use up limits fastest. As of October 2026, Claude Code's default model on Pro and Max is Opus 5.5, so unless you change it, every turn runs on Opus, renames and one-line fixes included. Anthropic's cost guide lists what else drives usage up in a long session:

• - Claude Code sends the whole conversation with every request, and again with each batch of tool results, so a short question late in a long session draws usage for all of it.
• - The first message after a break longer than the prompt-cache lifetime misses the cache and reprocesses the full context.
• - Subagents, workflows and agent teammates send their own requests. Agent teams use about 7× the tokens of a standard session when teammates run in plan mode.
• - Scheduled /loop tasks and goal check-ins fire while the session sits idle, sending the full context each time.

On a Pro, Max, Team or Enterprise plan, /usage also breaks recent usage down by skill, subagent, plugin and MCP server, and recommends a change for anything behind 10% or more of it.

## How do you make Claude Code limits last longer?

Match the model to the turn first, then keep the context small. Anthropic's cost guide says Sonnet handles most coding tasks well and costs less than Opus, and its guide to choosing a model says to pick a smaller model when the work is routine. How to send quick coding tasks to a cheaper model goes through which tasks move safely.

• - Make Sonnet your default with /model sonnet, and switch with /model opus when a task needs it.
• - Or use opusplan, which runs Opus in plan mode and Sonnet for execution: run /model opusplan, then press Shift+Tab to cycle to plan mode.
• - Give subagents a smaller model with model: haiku or model: sonnet in their frontmatter, or set CLAUDE_CODE_SUBAGENT_MODEL for subagents that don't name one. Otherwise a subagent runs on your session's model, so an Opus session runs Opus subagents.
• - Run /clear between unrelated tasks. Use /compact with instructions when you need to keep going.
• - Leave effort at its default for most tasks, which is medium on Opus 5.5, Sonnet 5.5 and Haiku 5.5, and lower it with /effort low for simple ones. Thinking tokens bill as output tokens.
• - Keep CLAUDE.md under 200 lines, move workflow instructions into skills, and turn off unused MCP servers with /mcp. /context shows what fills the window.
• - Press f in /usage to have Claude Code apply its recommendations (v2.1.305 or later).

A subagent for noisy, simple work, saved as .claude/agents/test-runner.md, looks like this:

```markdown
---
name: test-runner
description: Runs the test suite and reports only the failures
model: haiku
---
Run the tests. Return the names of failing tests and their error output, nothing else.
```

## How much does moving routine turns off Opus stretch a plan?

How far a plan stretches depends on how many of your turns are routine and where you move them. As of October 2026, Sonnet 5.5 costs half as much as Opus 5.5 at list price for every token type: $2 against $4 per million input tokens and $10 against $20 per million output tokens. Haiku 5.5 costs between a fortieth and a quarter of Opus 5.5 per token, depending on prompt size and caching.

Take the turn mix from AutoRouter's example week, an illustration on its landing page: 30% hard, 45% medium and 25% quick turns. With every turn the same size, and cost counted in Opus turns, list-price arithmetic gives these results:

In this mix, the medium turns decide how far a plan stretches. Moving only the 25 quick turns gets you at most about 30% more work per dollar, while moving the 45 medium turns as well nearly doubles it.

This is an illustration built on list prices. It describes usage credits exactly, since they bill at API rates, but for plan limits Anthropic says only that the model you choose affects how fast they run down; it doesn't publish the weights. Real turns also differ in size, so your mix of tokens matters as much as your mix of turns. The last two rows hold only if Sonnet does your medium turns well enough, and Does model routing make your results worse? looks at that trade-off.

## Can a tool pick the model for each Claude Code turn?

Yes. AutoRouter, the Mac app for Codex and Claude Code, picks a model for each turn. It shows up as a model in the Claude Code CLI: choose AutoRouter with /model and keep working. AutoRouter works out how hard each turn is, sends hard work to the strongest model your plans cover, and sends quick tasks to the cheapest capable model, which is often your plan or a low-cost open model. In Claude Code, Claude models run on your Claude plan. How AutoRouter picks a model for each turn covers the full order, including the plain-English rules you can write, such as sending git commands to a low-cost model. Every turn shows its model, what it cost and why. AutoRouter wrote this post.

AutoRouter's landing page also runs an illustrative example week of 30% hard, 45% medium and 25% quick turns, at Claude Opus, Sonnet and Haiku 5.5 list prices with cached input at a tenth of input. In it, AutoRouter Pro sends 85% of quick and 75% of medium turns to open models, which its included usage pays for. As the landing showed it on October 22, 2026, AutoRouter Free does 1.9× the work of using a frontier model for everything in that week, and AutoRouter Pro does 2.9×. Those figures describe an illustrative week and promise nothing about yours.

Choosing a specific model in /model sends turns straight to it; AutoRouter routes only while it is the model selected. AutoRouter is in beta and runs on macOS 14 or later, and Claude Code needs Node 22.18 or later to use it.

## What can you do when you hit the limit?

Waiting costs nothing, and every other option trades money or setup for time. The choices as of October 2026:

## Questions

### Does switching models get you past a Claude Code usage limit?

Switching models gets you past a model-specific limit only. The session and weekly limits are shared across all models, so /model doesn't help after You've hit your session limit or You've hit your weekly limit. After You've hit your Opus limit or You've hit your Sonnet limit, switching to a model outside that family keeps you working, though the next request re-reads the whole conversation without cache hits.

### Do claude.ai chats count against Claude Code limits?

Yes. Claude Code, claude.ai, Claude Desktop and IDE use draw from the same limits, so a long chat on the web leaves less for the terminal. The /usage breakdown counts only this machine's Claude Code sessions.

### Can you spend the Max plan's API credits in Claude Code?

No. The monthly API credits that come with Max can't be used in Claude Code or the Claude apps, including as extra usage. They are for building on the Claude Platform, and overflow in Claude Code bills to usage credits instead.

### Does /compact save usage?

Yes, /compact saves usage on the turns that follow it, but the compaction itself is a large request, because it reads the whole conversation to summarize it (cost guide). When you don't need continuity, /clear costs nothing.

### What happens in AutoRouter when your Claude plan runs out?

With AutoRouter Pro, Claude Code turns keep going from Pro's monthly included usage until your Claude plan resets. If that runs out too, AutoRouter asks you to top up, and credits you buy never expire. As of October 2026, Pro costs $20 a month with $20 of included usage, and new accounts get their first month at no charge, or until $20 of usage is spent, whichever comes first.

## Sources

• - Plans and pricing, Anthropic
• - What is the Pro plan?, Claude Help Center
• - What is the Max plan?, Claude Help Center
• - Use Claude Code with your Pro or Max plan, Claude Help Center
• - How do usage and length limits work?, Claude Help Center
• - Claude Fable models on your plan, Claude Help Center, September 2, 2026
• - Manage usage credits for paid Claude plans, Claude Help Center
• - Claude Code weekly limits promotion, Claude Help Center, September 14, 2026
• - What is a limit reset?, Claude Help Center, September 22, 2026
• - Higher usage limits and a SpaceX compute deal, Anthropic, May 6, 2026
• - Introducing Claude Opus 5.5, Anthropic, September 22, 2026
• - Manage costs effectively, Claude Code docs
• - Error reference: usage limits, Claude Code docs
• - Model configuration, Claude Code docs
• - Status line: rate limit usage, Claude Code docs
• - Wait for a usage limit to reset, Claude Code docs
• - Subagents: choose a model, Claude Code docs
• - Pricing, Claude Platform docs
• - Choosing a Claude model and effort level in Claude Code, Anthropic, July 7, 2026
• - Pricing, ChatGPT Learn, OpenAI