• Claude
  • Story
  • Nov 9, 2026

Claude Fable 5.2: twice Opus 5.5's price, for when Opus falls short

My take: Twice Opus 5.5's price per token, and Anthropic itself says to start on Opus. Fable 5.2 is for the long job Opus didn't crack, not for every day.

What to know

  • Anthropic released Claude Fable 5.2 on November 9, 2026. It is the same model as Claude Mythos 5.2, with safeguards in cybersecurity and biology that "will block most cybersecurity-related tasks", per Anthropic; Mythos 5.2 runs without them, for organizations vetted through its verification programs. Output is up to 30% faster than Fable 5.1, Anthropic says.
  • On the API, the model ID is claude-fable-5-2: $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1, and twice the price of Opus 5.5 ($4 and $20), per Anthropic's model table. A fast mode, up to 2.5 times faster, costs $16 and $80. It is also on AWS, Google Cloud and Microsoft Azure.
  • On FrontierSWE v2, a coding benchmark, Anthropic measures 69.9% for Fable 5.2 against 62.3% for Opus 5.5: its own figures, not an independent test.
  • On claude.ai, Fable is included up to half the weekly limit on Max, Ultra and premium Team seats, and Enterprise has it too, per the pricing page on November 16, 2026. On Pro, it runs on usage credits, after a launch offer that included it from November 9 to November 16, 2026. The free plan doesn't get it.

Spotted on @ninon.ia_officiel and @henriexploria.

What changes

Anthropic's own advice: start with Opus 5.5, which it calls its "strongest daily driver", and move up to Fable 5.2 for long autonomous work, or when Opus 5.5 at a higher effort level still falls short. In Claude Code, that means /model for one hard task, not a new default: every token costs twice as much on the API, and on Max, Fable stops at half your weekly limit. Pick a task where Opus stalled, run it again on Fable, and weigh the result against the bill.

Source: Anthropic, “Introducing Claude Fable 5.2 and Claude Mythos 5.2”, November 9, 2026

Drafted with Claude, Anthropic's AI, from my take. Reviewed and owned by me.

Up next