Opus 5.5 is the new default and Sonnet 5.5 is the fast option. Two launches in ten days make this the moment to audit your CLAUDE.md.
Two model launches landed in ten days. Opus 5.5 is the new default in Claude Code, and Sonnet 5.5 is the quick option for well-defined work.
| Model | Speed | Cost and tokens | Best for |
|---|---|---|---|
| Opus 5.5 | Over 30% faster than Opus 5 | Costs 40% less | Most work; lands at roughly Fable 5.1 level |
| Sonnet 5.5 | Over 30% faster than Sonnet 5 | Far fewer tokens per task | Well-defined tasks, like a bug fix with a clear repro |
Opus 5.5 also writes shorter replies, with far fewer "Claudisms" and em dashes, and sticks to what you actually asked for. Switch to Sonnet with /model sonnet.
Re-run prompt-audit
New models are the moment to check your instructions. Since v2.1.283, run /checkup prompt-audit (or /doctor prompt-audit; /claude-api prompt-audit still works).
It reads your CLAUDE.md files, skills, agents and custom commands, then flags:
- Instructions written for older models
- Stale file paths
- Rules that contradict each other
It tells you what to change or delete. Nothing is applied until you say so: findings go to a report (PROMPT_AUDIT.md) and edits to a patch file (prompt-audit.patch).
Two setup changes worth knowing
- AGENTS.md now counts (v2.1.277). If your repo has an AGENTS.md and no CLAUDE.md, Claude Code uses it as your project instructions. To load both, open
/config, then Project instructions. Since v2.1.281 this also works on Bedrock, Vertex and Foundry. - claude.ai skills and plugins now sync to the terminal (v2.1.275). Signed in to the CLI with your claude.ai account? The skills and plugins you enabled there now load in terminal sessions too. Opt out with
"syncClaudeAiSkills": falseor"syncClaudeAiPlugins": false.
Build Faster.
Earn Smarter. Stress Less.
See how AI can help your business communicate better with your customers
Start now
Lorem ipsum dolor sit amet consectetur
No items found.
