← Other blogs

Opus limit in Claude Code: five ways to keep working when capacity drops

Switch to Sonnet, put subagents on Haiku, clear between tickets and watch your reset clocks locally.

· 2 min read

Running out of the Opus limit in Claude Code usually happens right before tests pass. When the warning appears, or right after the error blocks your next prompt, five fast adjustments keep you productive without abandoning your branch.

Five fixes when Opus runs low

  1. Switch models with /model. According to Claude Code's cost guide, hitting the model-specific "You've hit your Opus limit" message still leaves other model families available. Switching to Sonnet lets you finish implementation right away, whereas hitting the shared five-hour session or weekly cap blocks all models.
  2. Put subagents on Haiku. Subagents inherit your main session's model by default. Setting model: haiku in subagent configuration keeps doc fetches, log scans and test runs off your Opus quota.
  3. Clear between unrelated tasks. Use /rename if you want to return later, then run /clear so giant diffs and file reads from your previous ticket stop riding along on every prompt.
  4. Compact with a target. When you must keep thread continuity, run /compact with a short instruction such as focusing on test output and edited files instead of keeping full raw logs.
  5. Lower reasoning effort. Extended thinking bills as output tokens; running /effort at a lower tier trims deep-thinking overhead on routine edits.

Why one status check is not enough

Metric What it tells you Where to read it
Live session tokens & attribution What burned recent quota on this machine /usage inside Claude Code
Five-hour & weekly reset clocks Remaining capacity and local reset time UsageAtlas desktop view
Threshold alert Early warning before a window hits zero One desktop alert per cycle in UsageAtlas

UsageAtlas tracks supported Claude Code windows alongside Codex, Cursor, OpenCode, GitHub Copilot and other local coding tools, free and open source, without needing an account for local tracking or reading your prompts and responses.

Keeping a buffer for the work that needs Opus

Treat Opus like a specialist rather than a default. Run planning or tricky architectural steps on Opus, drop to Sonnet for writing code and tests, and set a desktop threshold alert (for instance at 20% remaining) so you notice the dip before the prompt locks. If an urgent deploy cannot wait for a reset window, /usage-credits opens pay-as-you-go extra usage on eligible subscriptions. Teams and Enterprise organizations can also set individual or shared usage-credit spend caps in admin settings, and running /insights writes a local HTML report summarizing friction points and session habits over up to 200 recent sessions.

FAQ

What is the difference between /clear and /compact for Opus limits?

/clear resets the conversation context cleanly for zero compaction cost, while /compact spends one summarization turn to compress earlier context so you can keep working in the same thread.

Will /usage show usage from my other laptop?

No. Anthropic notes that /usage attribution figures are computed from local session history on that machine, so claude.ai and other devices are excluded.

Is UsageAtlas only for Claude Code?

No. It tracks local usage across multiple coding assistants including Claude Code, Codex, Cursor, OpenCode, GitHub Copilot and Gemini CLI in one place.