#fable

Working with Fable 5.2, the session gets compacted into a summary after almost every task, and the app shows how many tokens each compaction saved. Put rules in files and give each task in one message so nothing gets lost.

Anthropic's Claude Fable 5.2 is up to 30% faster, costs $8 per million input tokens and scores 49.1% on FrontierCode 2.0, though Opus 5.5 remains the everyday pick. Includes a table for picking the right model.