You're mid-task and Claude Code stops with You've hit your session limit. Here's what's going on, what you can do right now, and how to make the next window last longer.

1. Find out which limit you hit

On a Pro or Max plan, every request counts against two limits at once: a five-hour session limit and a weekly limit. Whichever runs out first stops you. Both are shared with claude.ai and Claude Desktop, so a long chat on the web leaves less for the terminal.

Run /usage to see both limits and when each one resets. The message also tells you which one you hit:

  • You've hit your session limit: the five-hour window. It resets in full five hours after your first message.
  • You've hit your weekly limit: resets at a fixed day and time on your account (Settings > Usage shows it).
  • You've hit your Opus limit / Sonnet limit: a model-specific cap. This is the only case where switching models gets you going again.

A 529 overloaded error isn't a usage limit at all. It means the servers are busy, so retry or switch models.

2. What you can do right now

Option How Cost
Wait Claude Code picks the task back up after the reset in interactive sessions (v2.1.234+) Nothing
Use a saved limit reset Settings > Usage on the web or in Claude Desktop, when Anthropic has given you one Nothing
Turn on usage credits /usage-credits opens Settings > Usage. Set a monthly cap API rates per turn
Use an API key Sign in with a Claude Console account Per token, billed apart from your plan
Upgrade Max or Ultra $100 or $200 a month as of November 2026. Weekly limits still apply

If you just need to finish one thing, usage credits with a low cap is the least painful. Every overflow turn bills at its model's API price, so leaving Opus on for a rename gets expensive.

3. Make the next window last

Most people burn their limit on the model, not the work. Claude Code's default model on Pro and Max is Opus 5.5, so every turn runs on Opus unless you change it: renames, commit messages and one-line fixes included. Anthropic's own cost guide says Sonnet handles most coding tasks well for less.

  • Make Sonnet your default: /model sonnet. Switch to /model opus when a task really needs it.
  • Or use opusplan: /model opusplan runs Opus while you plan and Sonnet while it executes (Shift+Tab cycles into plan mode).
  • Give subagents a smaller model: add model: haiku or model: sonnet to their frontmatter, or set CLAUDE_CODE_SUBAGENT_MODEL. Otherwise an Opus session runs Opus subagents.
  • /clear between unrelated tasks. Claude Code sends the whole conversation with every request, so a quick question at the end of a long session pays for all of it. /compact helps when you need to keep going, but the compaction itself is one big request.
  • Watch what runs while you're away: scheduled /loop tasks, subagents and agent teams all send their own requests. Agent teams can use about 7× the tokens of a normal session.
  • Keep usage on screen: /usage breaks recent usage down by skill, subagent, plugin and MCP server, and flags anything behind 10% or more of it.

The model swap is the big one. Moving routine turns off Opus stretches a plan much further than trimming context does.

4. Or let something pick the model for you

Switching models by hand works until you forget. I'm building AutoRouter, a Mac app that does it per turn: a router that lives inside all your coding agents and runs on the AI subscriptions you already pay for. In Claude Code you pick it with /model. Hard work goes to the strongest model your plans cover, and quick tasks go to the cheapest model that can handle them. Every turn shows which model ran, what it cost and why.

When one plan hits its limit, AutoRouter moves to the connected account with the most room left before it spends any credit. With Pro, turns keep going from its monthly included usage until your plan resets. It's free to start and runs on macOS 14 or later.

Quick checklist

  1. /usage: which limit, and when does it reset?
  2. Model-specific limit? Switch models. Session or weekly? Wait, use a saved reset, or turn on capped usage credits.
  3. Next session: /model sonnet (or opusplan), smaller models for subagents, /clear often.

For the same breakdown across every coding agent and plan, see AI coding usage limits, explained.

Disclosure: I'm a co-founder of AutoRouter.