Getting the Best Out of Fable 5: Effort, Thinking & Prompting Tips

Fable 5 follows instructions more literally and reasons more autonomously than earlier Claude models. A few habits make a big difference.

1. Front-load the full task

Fable 5 shines on long-horizon work when it gets the complete specification up front — goal, constraints, and what "done" looks like — in one well-written turn. Dripping requirements across many turns wastes tokens and degrades performance compared to a single clear brief.

2. Use effort as your main dial

The effort parameter (low → max) replaces the old habit of tweaking temperature or thinking budgets:

• xhigh — coding and agentic work; the sweet spot for hard tasks
• high — the default; right for most intelligence-sensitive work
• low/medium — fast, scoped tasks where latency matters

Higher effort up front often reduces total cost on agentic work: better planning means fewer wasted tool calls.

3. Let adaptive thinking do its job

Enable thinking: {type: "adaptive"} and stop micromanaging. Fable 5 decides per-request whether deep reasoning is warranted. Don't try to disable thinking explicitly — on Fable 5 that returns an error; just omit the parameter if you don't want it.

4. Drop the aggressive scaffolding

Prompts written for older models — CRITICAL: you MUST..., "after every 3 tool calls, summarize" — overtrigger on Fable 5. It follows polite, precise instructions faithfully. State what you want once, clearly.

5. Be prescriptive in tool descriptions

Fable 5 reaches for tools deliberately rather than reflexively. If you build agents, write tool descriptions that say when to call the tool ("Call this whenever the answer depends on current prices"), not just what it does.

Want to experiment without writing code? Open the playground and test these patterns directly against claude-fable-5.