Anthropic is reportedly preparing a new model with stronger reasoning and intuition, able to work out what users actually want even when their instructions are incomplete. The Information reported on October 19, citing people familiar with the work. The report gives neither a name nor a release date.
The name circulating is Fable 5.5, but the evidence is indirect. For a short time, Microsoft Azure's catalog in the East US 2 region listed an identifier "claude-fable-5-5" as generally available; the entry was later removed. The same name also appeared in a proposed change to the open-source project LiteLLM, which tracks model pricing. The listed price, $10 per million input tokens and $50 per million output tokens, matches Fable 5.1 exactly, so it may be a copied placeholder.
Fable 5 Subscription Guide: Plans, Credits, Pricing & Limits | Blog ...
On the official side, Anthropic's model page still lists Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5. Anthropic has not confirmed Fable 5.5, its price or an API. Fable 5.1's documentation still presents it as the choice for demanding reasoning.
A tester cited by The Information describes the model's main capability as goal inference: given a bug, it would not just fix it but identify the broader problem and propose a comprehensive fix. If confirmed, the practical difference would be concrete: less need for elaborate prompts and more interpretation handled by the model itself.
At the same time, The Information reports that Google is working on Gemini 4 Carbon and that OpenAI is preparing a mid-tier model. Nothing is official yet.
No comments yet. Be the first!