Claude Fable 5 is back. The US government pulled it three days after launch, then lifted the ban, and it is available again in Claude Code. It is Anthropic's most capable model, built for your most ambitious work: it investigates your codebase before it acts, uses relevant context as it works, and checks its own work before saying it is done. Here is how to get the most out of it, based on the official guidance from the Claude team.
Switch it on
In Claude Code, type /model and pick Claude Fable 5. Through 7 July it is included in your subscription (up to 50% of your weekly usage limits), then you can keep using it with usage credits. Pricing is $10 per million input tokens and $50 per million output.
1. Give it the goal, not the steps
Describe the outcome and let Fable 5 plan the path. Use /goal <your outcome> to keep it working until a completion condition is met, instead of spelling out every step yourself.
2. Hand off big, asynchronous work
Give it tasks you would normally break up. It holds multi-hour, even multi-day sessions without losing the thread, so you can size up a task and walk away.
3. Skip the guardrail prompts
You do not need to add "make sure to test." Fable 5 reviews its own work with less prompting. Qualitative feedback (like "this feels over-engineered") works better than itemized instructions.
