Why Claude Code uses up your limit so fast

Claude Code uses up your limit fast because every request carries your whole conversation, so a long session makes every later request bigger. That is the first reason Anthropic gives. Its list also includes cache misses after a pause, subagents and workflows, scheduled tasks that fire while you are idle, compaction, and your choice of model and effort level. These are documented behaviours of the tool, and most of them are things you can steer.

Anthropic does not publish token or message counts behind any limit, so nothing here can tell you how much a given action costs in your plan. What follows is what Anthropic documents, then, separately, what users have reported.

What Anthropic documents

The list below comes from the Claude Code costs documentation and the usage and limits article, as of November 2026.

Long context is resent with every request

Claude Code "sends your full conversation with every request", and each tool use sends another request carrying the results. Prompt caching makes that history cheaper to re-read, but "a one-line question in a session that has been open all day still draws usage for the whole conversation" (costs documentation). Anthropic's tip: "Stale context wastes tokens on every subsequent message." A session that has wandered through three unrelated tasks carries all three on every new message.

Cache misses after a break

The cache only helps while it is warm: "your first message after a break longer than the cache lifetime misses the cache and reprocesses your full context", which Anthropic lists among the reasons usage climbs in a long session. Switching models has a similar effect: each model has its own prompt cache, so after a switch "the next request re-reads the whole conversation with no cache hits" (error reference).

Subagents, workflows and agent teams

Work you hand to subagents still counts: "The subagent's own requests still draw on your usage." Anthropic lists subagents and workflows, agent teammates and goal check-ins as sources of usage. Agent teams use about 7x more tokens than standard sessions when teammates run in plan mode. Anthropic's own example of a burst that can empty the week is "a large workflow fanout" (error reference).

Things that run while you are idle

Anthropic lists several things that can spend usage while a session sits idle: a session-scoped scheduled task (such as a /loop) "fires on its interval even while the session is idle, sending your full context each time"; messages from your other sessions and goal check-ins can start new turns; and "Session recap and idle compaction can use tokens while you're idle." Prompt suggestions also send a short request after each response (skipped when you are close to the limit). Any of these can explain a limit that moved while you were away from the keyboard.

Compaction is itself a large request

/compact "reads the conversation it summarizes, so compacting a large context is itself a large request". When you want a fresh start rather than continuity, "/clear costs nothing." The trade-off is the subject of /compact vs /clear.

Model choice

"Opus costs several times more per turn than Sonnet, and Sonnet more than Haiku. Spending Opus on routine work is the fastest way to drain a daily limit" (Help Center). Fable models "use [your limits] faster than other Claude models" (Fable article). As of November 2026 the default model on Pro, Max, Ultra, Team and Enterprise is Opus 5.5, so you are on the more expensive end unless you change it (model configuration docs). Which one costs least is covered in which Claude model uses the least of your limit.

Effort level

"Higher effort uses more tokens" (Help Center). Set it with /effort or claude --effort. Thinking cannot be turned off on Opus 5.5, Sonnet 5.5, Haiku 5.5 or Fable models, so lowering effort is the lever there (costs documentation).

Everything else you attach

The general best-practices article also names message length, file attachment size, tool usage, artifacts and multi-step tasks (best practices), and the usage limits article notes that "Tools and connectors are token-intensive" (Help Center). In Claude Code, MCP servers add to your context, although their tool definitions are deferred by default until Claude uses a tool. Running /context shows what takes space.

Other ways your limit goes down

• Shared with chat. Claude, Claude Desktop and Claude Code draw on the same allowance, so conversations elsewhere count. See do Claude chat and Claude Code share limits.
• Two limits at once. Use counts against the session and weekly allowances together, so a big burst can drain the week. See session limit vs weekly limit.
• Non-interactive runs. claude -p and the Agent SDK draw from your plan when signed in with one (Help Center).

What users report (not official behaviour)

These are reports on GitHub. They are real posts from real users, but they are not Anthropic's description of the product, and nothing published confirms whether any of them is a defect.

• Users have reported limits feeling faster than before or running out early, including on Sonnet 5.5 in November 2026 (#108661) and on a Max plan since mid-October 2026 (#105621).
• Users have reported usage rising with no activity (#94581, #104614).

Anthropic's idle-usage list above (scheduled tasks, recap, idle compaction) is the documented place to start if you see that. Anthropic also says it may "adjust usage limits at our discretion to manage capacity" (Help Center), and it has changed Claude Code's limits several times this year (for example, doubling the five-hour limits on May 6, 2026). If your usage changed and your habits did not, a limit change on Anthropic's side is possible, not provable from outside.

How to find out where yours went

Run /usage in Claude Code. On a paid plan it "includes a breakdown of what counts against your plan limits": skills, subagents, plugins and MCP servers as percentages, recommendation rows for behaviours behind at least 10% of recent usage, and your heaviest /loop tasks (commands, costs docs). Keep in mind that the breakdown is computed from session history on this machine, so usage from other devices or claude.ai is not included, although the limit bars are account-wide.

Once you know the cause, Anthropic's own tips to make Claude Code limits last turns it into habits.

Watching it climb

Headroom is a free Mac menu bar app that shows session and weekly use for each Claude Code account signed in on your Mac. It checks every five minutes, so when you come back to your Mac you can see whether the numbers moved while you were away, which helps spot idle usage. It does not break usage down by cause, so /usage is still the tool for that. Headroom never sends prompts, and it reads usage figures Anthropic does not publish as an official interface, so it can break until updated. It is independent of Anthropic and not endorsed by it.