Getting the most out of Opus 5.5 in Claude and Claude Code
Anthropic explains how to prompt, steer, and check long Claude Opus 5.5 runs in Claude and Claude Code.
Anthropic’s Claude blog says Opus 5.5 thinks before every reply, continues longer on multi-step coding than Opus 5, and reports plainly what it did. The guide advises giving the full task and a clear finish line, removing “think carefully” instructions, and adding mid-run direction in Claude Code. It recommends CLAUDE.md rules for when to keep going versus stop before destructive actions. For large audits, it suggests fanning work out to subagents and keeping a TASKS.md checklist because long runs are later summarized.
- Opus 5.5 thinks before every reply and sustains longer multi-step coding than Opus 5.
- Give the whole task and a finish line, and drop think-carefully prompt lines.
- CLAUDE.md should say when to continue and when to stop before destructive actions.
- Large audits can be split across subagents, with each result’s evidence checked.
- Keep a TASKS.md checklist so progress survives context summarization.
- Vendors
- Anthropic
- Products
- ClaudeClaude Code
- Organizations
- Anthropic
- AI models
- Claude Opus 5.5Claude Opus 5
From the original article
10 min read at claude.dev
Opus 5. 5 works well with the way you already use Claude. A few things behave differently, though: it works for longer on its own, it tells you plainly what it did, and it thinks before every reply.
By saikatsg · claude.dev
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.