Opus 5.5
Claude is finally back.

TL;DR
- Claude's Opus 5.5 is now the top-ranked AI model, with new versions releasing frequently.
- Prompting Claude Opus 5.5 requires different techniques than older versions; 'think harder' commands should be removed.
- Instead of vague instructions, specify exact design patterns to avoid and name the desired elements.
- Empower Claude to explore connected applications and files before acting, and clearly define the 'finish line' for long tasks.
- Use tags to differentiate pasted text from your own instructions, and leverage visual inputs like screenshots and charts.
- Claude can now effectively check work, catching errors in dates, data, and formulas.
- Basic prompting principles like specifying scope, length, and using action verbs remain crucial.
- The 'Medium' effort level is recommended for most tasks, with 'Extra' for highly complex ones.