Claude Code usage draining fast? What Anthropic says

What Anthropic staff said about Claude Code limits

The complaints peaked on Oct 21 and 22. One post that drew more than 200 replies said usage was "draining 5x faster" (reported, not confirmed). Anthropic staff answered in the replies rather than with an announcement:

"hey, our limits haven't changed, but could you run /feedback in a session where you noticed it draining faster and dm me the feedback id? I can take a look"

Lydia Hallie, Claude Code team at Anthropic, Oct 22, 4:26 PM ET (post 2113366380842254746)

"No changes on our end. Can you run this command and paste the full output here? Happy to help debug. claude -p /usage | pbcopy"

Boris Cherny, who works on Claude at Anthropic, Oct 22, 1:34 AM ET (post 2113141937645769087)

"No changes on our end, we monitor usage really carefully. To debug, run /usage to see a detailed token by token breakdown for your own usage. Most often it’s a runaway loop or really inefficient skill"

Boris Cherny, Oct 4, 5:50 PM ET (post 2106864506894970959)

So there is no confirmed limit cut and no acknowledged bug. That doesn't mean nothing is wrong with your account. It means Anthropic wants the session data to check it.

How to see what is draining your Claude Code usage

1. Run /usage. On Pro, Max, Team and Enterprise it shows your plan bars plus a breakdown of recent usage by skill, subagent, plugin and MCP server. Press d or w to switch between the last 24 hours and 7 days.
2. Read the recommendations. Any habit behind 10% or more of your recent usage gets its own row, such as "Trim CLAUDE.md" or "Compact before you step away". Press f to let Claude Code apply the ones it can.
3. Check the cache line. "Prompt cache (main)" in /usage shows cache misses and whether the cache is warm. Several misses in a long session is a common reason a quiet afternoon costs a lot.
4. Send it to Anthropic. Staff ask for /feedback in the session that drained, or the output of claude -p /usage | pbcopy.

The breakdown only covers sessions on that machine. Usage on claude.ai or another computer counts against the same limit but won't show up there.

Why Claude Code uses more than you expect

Anthropic's own Claude Code cost docs list these, and the staff replies point the same way:

• Long context: Claude Code sends your whole conversation with every request, so a one-line question in a session that has been open all day still pays for the full history.
• Cache misses: your first message after a break longer than the cache lifetime reprocesses everything. Boris Cherny said interactive sessions default to a 1-hour cache, and API workloads to 5 minutes.
• Skills, plugins and loops: "Most often it’s a runaway loop or really inefficient skill," Cherny said. In one thread he told a user to disable a plugin that was using most of their tokens.
• Subagents and agent teams: each subagent sends its own requests. Agent teams use about 7x more tokens than a normal session when teammates run in plan mode, per the docs.
• Compaction: /compact reads the whole conversation it summarizes, so compacting a huge context is itself a big request. /clear costs nothing.
• Thinking and effort: thinking tokens count as output. You can't turn thinking off on Opus 5.5, Sonnet 5.5 or Haiku 5.5, but /effort lowers it.
• Background turns: scheduled tasks, goal check-ins and messages from your other sessions can start a turn while you're away, and each one resends your context.

Fixes that cut Claude Code usage

• /clear between unrelated tasks instead of keeping one session open all day.
• Compact earlier. Lydia Hallie suggested /autocompact 400k on 1M-context models, so messages never carry close to a million tokens.
• Let idle compaction work. Since mid-October Claude Code compacts long idle sessions just before the cache expires, so coming back later only reprocesses the summary (Hallie, Oct 14).
• Pick the model for the job. The docs suggest Sonnet for most coding and Haiku for simple subagents. Opus 5.5 already runs about 25% further within limits than Opus 5, per Anthropic.
• Trim what loads every time: keep CLAUDE.md under 200 lines, move workflow notes into skills, and disable MCP servers you aren't using with /mcp.

Out of Claude Code usage right now?

The limit message tells you when the 5-hour or weekly window resets. Until then, Pro and Max can turn on usage credits with /usage-credits (billed at API rates, with a monthly cap you set). If it happens every week, Max 5x is $100 and Max 20x is $200 a month vs $20 for Pro (Anthropic pricing, checked 2026-10-23). See Claude usage limit reached for each option.

Did Anthropic reset Claude limits?

Not for this. The last limit reset Anthropic posted was "Opus 5.5: saved reset for Pro, Max, Team; 5-hour limits up 20%" on Sep 22, 12:44 PM ET (source). Every Claude reset and limit change is in the Claude reset and limit change log, and the next one will show on when does Claude reset within the hour.