Get more out of Opus 5.5
Hey, it's PT.
Here is how to get more out of Opus 5.5.
What you get: which model to pick for each job, which effort level to run, and how to make your limits last longer. Plus 1 command that updates your setup for the new model.
How to use this page: keep it open next to Claude and check it before you move an old prompt over.
The quick way: 1 command updates your CLAUDE.md and skills for Opus 5.5, so your old setup works with the new model. Type it once in Claude Code and you are set.
👉 Jump to the update command
Which model for which job 🧭
Scores are from Anthropic's launch table.
| The job | Use | Why |
|---|---|---|
| Coding and code review | Opus 5.5 | 57.8% vs Fable's 51.8% on CursorBench |
| Docs, spreadsheets, decks | Opus 5.5 | 1846 vs Fable's 1735 on real office work |
| Driving your computer | Opus 5.5 | 81.8% vs Fable's 80.7% |
| Business automations | Opus 5.5 or Astra | 41.4% for Astra vs 40.0%, close enough to call even |
| Science research agents | GPT 6 Astra | 64.6% vs Opus 5.5's 58.7% |
| The hardest, hours long work | Fable 5.1 | Anthropic's pick when Opus 5.5 at xhigh falls short. Costs 2.5x per token |
Which effort to run 🎚️
Effort is how hard it thinks, and it is the biggest dial on your limit.
| Effort | Use it for |
|---|---|
| Low | Quick edits and questions. Often close to medium for much less. |
| Medium | The default. Right for most work, and it matches Opus 5 at high. |
| High | A bug that won't die, or a big change across many files. |
| xhigh / max | Only when a lower level fell short on that exact task. These drain limits fast. |
To set it in Claude Code, type /effort medium (or low, high, xhigh, max). Type /effort on its own for a slider. Switch models with /model opus or /model fable. An effort level saved from an older model doesn't carry over, so set it once.
Make your limits last ⏳
- More room than you think. In Claude Code your 5 hour limits went up 20%, and because Opus 5.5 is cheaper, the same limit gets you 25% more work.
- Save Fable for the jobs in the table. For everyday work it spends 2.5x the tokens. Lower the effort before you switch models.
- Hold your free reset for a crunch. Pro, Max and Team get one. Find it in Settings, then Usage, and click Reset for free. It can't be undone, so spend it mid job when you're about to hit the wall.
Prompt it the new way ✍️
Look around first. When a task spans your email, docs and spreadsheets, say:
Before you change anything, open everything that could be
relevant, including things I didn't mention.A finish line. For long jobs it pauses to report progress. Keep a to do list in a file and say:
Keep going until every item is done.Stop the rethinking. In a long chat it rechecks earlier replies, which slows each new answer. Say:
Treat earlier answers as done unless I ask about them.Update your setup once 🔄
Do this once. Type this in Claude Code. It updates your CLAUDE.md and skills for the new model.
/claude-api prompt-auditFull detail: Prompting Claude Opus 5.5 and Getting the most out of Opus 5.5.
Keep going 🚀
More tips and prompts, on Instagram and straight to your inbox.
Rather learn it live? The AI Workflow Workshop is 2 hours, hands on, no coding required. You build 1 repeatable workflow on your own work.
Here's what happens in the 2 hours:
- Set up the file your AI reads before anything else.
- Build one repeatable workflow against your real work.
- Run it live, watch it miss, and teach it to correct itself.
- Ask about your own use case.
