Learn how to use recalibrated effort levels, adaptive thinking, and new defaults to optimize your Claude Code setup with Opus 4.7.

Claude

Anthropic shares best practices for using Claude Opus 4.7 within Claude Code, covering key behavioral changes from Opus 4.6. Key guidance includes structuring interactive sessions by front-loading task context, reducing back-and-forth turns, and using auto mode for long-running tasks. A new 'xhigh' effort level is now the default, sitting between 'high' and 'max' to balance reasoning quality and token cost. Fixed thinking budgets are no longer supported; instead, Opus 4.7 uses adaptive thinking, deciding when to reason deeply based on context. Other notable changes include less verbose default responses, fewer tool calls in favor of more reasoning, and more conservative subagent spawning — all of which may require prompt adjustments for users migrating from Opus 4.6.

Best practices for using Claude Opus 4.7 with Claude Code