Claude Code Usage Limits: How to Check and Stretch Them

Claude Code usage limits are your Claude plan’s limits: on Pro, Max, Team and seat-based Enterprise, Claude Code draws on the same five-hour session limit and weekly limit as Claude on the web, Desktop and Cowork. Run /usage in a session to see both bars, when each resets, and which skills, subagents and MCP servers used the most. When a limit stops Claude mid-task, current versions wait in the open session and continue on their own after the reset; /usage-credits lets you pay to keep going, and /model helps only when the message names Opus or Sonnet. To make the allowance last, pick the model and effort per task and keep the conversation small. On an API key there are no plan limits at all: you pay per token.

How the plans compare, what the weekly reset is and what uses the allowance fastest across all of Claude are in Claude usage limits explained. This post stays inside the terminal.

Which limits apply to you

• Pro and Max: a five-hour session and a weekly limit, shared with the Claude apps. Max 5x and Max 20x give five or 20 times Pro’s per-session allowance.
• Team: per-member limits, 1.25 times Pro per session on a Standard seat and 6.25 times on a Premium seat, with a weekly limit on both.
• Enterprise: per-seat limits on seat-based plans; none on usage-based plans, which bill consumption at API rates.
• Claude Console API key, Bedrock, Agent Platform or Foundry: no plan limits and nothing to wait for. Spend is capped by workspace spend limits or your cloud’s budget controls.

One trap catches subscribers every week. Anthropic’s help article for Pro and Max↗ (opens in a new tab) warns that if an ANTHROPIC_API_KEY environment variable is set, Claude Code uses that key instead of your subscription, and you are billed at API rates rather than drawing on the plan. If your usage bars never move, check for the variable.

Check your usage with /usage

/usage, with the aliases /cost and /stats, is the one screen to learn. It opens while Claude is still working. According to the cost documentation↗ (opens in a new tab), it has three parts on a subscription:

• The Session block: tokens by model and an estimated cost. It is meant for API users; on Pro or Max the dollar figure is not what you are billed.
• Plan usage bars: how much of the session and the weekly limit you have used, and when each resets.
• A breakdown of recent usage by skill, subagent, plugin and MCP server, flags for any behavior such as long context or cache misses that accounts for 10% or more, and rows for the heaviest /loop or scheduled tasks. Press d or w to switch between the last 24 hours and the last 7 days. It covers this machine only, not other devices or claude.ai.

Anthropic’s pages disagree on one detail. The help article above says to monitor your remaining allocation with /status; in the current Claude Code command reference, /status shows version, model, account and connectivity, and the plan limits are on /usage. Use /usage. In the VS Code extension the same shares appear in the Account & usage dialog, and the Desktop app has a usage ring next to the model picker.

To watch the limits without running a command, add them to your status line. For Pro and Max subscribers, the JSON Claude Code passes to the script includes rate_limits.five_hour and rate_limits.seven_day, each with used_percentage and resets_at; they appear after the first response in a session. The Claude Code status line covers the setup.

#!/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 "session ${five:-?}% | week ${week:-?}%"

The limit messages, and what each one means

Before a window runs out, Claude Code can warn you with a line such as “You’ve used 85% of your session limit · resets 3:45pm”. When it does run out, you see one of four messages, listed in the error reference↗ (opens in a new tab):

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 or weekly limit: shared across all models, so /model does not bring you back. Wait, or buy more.
• Opus or Sonnet limit: applies only to that family. Switch with /model to a model outside it and keep working. The new model has its own prompt cache, so the next request re-reads the whole conversation with no cache hits.
• “Usage credits required for 1M context” is not a quota at all. It is an entitlement check on a [1m] model variant your plan only includes through usage credits, and it can appear with capacity left. Pick the variant without [1m] in /model.

What Claude Code does at the limit

From v2.1.234, when a session or weekly limit stops a task in an interactive session signed in with a subscription, Claude Code waits and then continues by itself. The bottom line reads “Usage limit reached · continuing automatically at 3:45pm · esc to cancel”. At the reset it sends Claude a fixed prompt to pick up where it stopped, rather than resending your last message. The details, from the interactive mode documentation↗ (opens in a new tab):

• Keep the session open. Exiting ends the wait, and it does not restart when you resume.
• The continued turn still asks for permissions as usual, so an unattended task can stop on a prompt. What runs without asking is set by your permission mode; see auto-approve in Claude Code.
• If it hits the limit again, it re-arms at most twice in a row and then stops.
• It does not start on its own for a reset more than 24 hours away, which a weekly limit often is, or in -p runs and background sessions. /rate-limit-options offers the wait, usage credits or an upgrade.
• To turn it off, switch off Continue automatically at usage limit in /config, or set autoContinueAtUsageLimit to false in your user settings.

To keep going now instead, /usage-credits (formerly /extra-usage) opens Settings > Usage on claude.ai for Pro and Max, where you turn on usage credits, billed at API rates. On Team and Enterprise without billing access, it sends your admins a request. /upgrade opens the plan page. One cost to know: the prompt cache lasts an hour on a subscription but drops to five minutes once you are drawing on usage credits.

Stretch the allowance: model and effort

The biggest lever is the model. On Pro, Max, Team and Enterprise, the default setting now resolves to Opus 5.5; before v2.1.280, Pro and Team Standard started on Sonnet 5. Anthropic’s cost guide says Sonnet handles most coding tasks well and costs less than Opus, and to keep Opus for complex architecture and multi-step reasoning.

• /model sonnet for routine edits. In the picker, press s on a row to switch for this session only; otherwise the choice becomes your default.
• /model opusplan runs Opus in plan mode and Sonnet for the edits.
• model: haiku in a subagent that only searches or summarizes, since subagents draw on your limits too.
• Fable models are not the default anywhere. On Max and premium seats they can use up to half of your weekly limit and draw on it faster; on Pro and standard seats they bill to usage credits, and Claude Code asks before it does.
• /effort low or medium on simple work. You cannot turn thinking off on Opus 5.5, Sonnet 5.5 or Fable, so effort is the dial there. The model configuration guide↗ (opens in a new tab) lists the levels.

Which model suits which job is in Claude Sonnet vs Opus.

Stretch the allowance: compaction and context

Every request carries the whole conversation, and on current models the window is 1M tokens, with automatic compaction at about 967K by default. A session can therefore grow very large before anything trims it, and every turn pays to re-read it. The /usage behavior flag for long context is the sign.

• /clear between unrelated tasks. It costs nothing, and /resume brings the old conversation back.
• /compact keep the failing test names and the plan when you need continuity. Compacting reads the whole conversation, so it is a large request itself; do it at a natural break, not every few turns.
• /autocompact 500k (v2.1.221 or later) sets a lower threshold, so long sessions are summarized sooner and each request stays smaller.
• After a long break on Pro or Max, accept the offer to resume from a summary instead of the full history.

Cutting tokens in general, from subagents to MCP servers to a shorter CLAUDE.md, is covered in how to reduce Claude Code token usage, and it is not repeated here.

Leave a note before the window closes

The 85% warning is a good moment to write down where things stand. Automatic continue only helps if the session stays open; a weekly limit days away, a closed laptop or a teammate taking over all start from a fresh session. If the work is tracked on fenbs, ask Claude at the warning to update the task: rewrite its plan with what is done and what is next, and add a comment with the files touched and the failing test. The next session reads it with fenbs_get_context and fenbs_get_item, and the History shows what the assistant changed and when. fenbs does not see your Claude plan or its reset times. The rest of that rhythm is in a task-tracking workflow for Claude Code.