The Library Media — Science, Technology & Health · AI News & Intelligence Desk
One Model, Two Doors: Anthropic Ships Claude Fable 5.2 for Everyone and Mythos 5.2 for the Vetted Few
By Maya Wilson
Follow The Library Media
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, November 9, opening access to the same underlying model through two sharply different doors: a broadly available version with stronger safeguards and a restricted version for organizations cleared to handle more sensitive work.
The distinction, Anthropic says, is not a difference in the underlying model itself. Fable and Mythos are the same model, with different access and safety arrangements. In Anthropic’s description, Fable includes strong cybersecurity and biology safeguards, allowing it to be made available to all users.
Fable 5.2 is available through the Claude API and Anthropic’s apps, as well as Claude Code, Amazon Bedrock, Google Cloud and Microsoft Foundry. Mythos 5.2, by contrast, is restricted to organizations approved through Anthropic’s Life Sciences and Cyber Verification Programs.
The split is the central design choice of the release: one underlying system made widely available with additional safeguards, while access to the less-restricted configuration is controlled through organizational verification.
Anthropic priced the new model at $8 per million input tokens and $40 per million output tokens, a 20% reduction from the $10 and $50 pricing of version 5.1. An optional fast mode, Anthropic says, can generate responses at up to 2.5 times the standard speed, priced at $16 per million input tokens and $80 per million output tokens.
The company also claims improvements in coding, reasoning and vision, along with outputs that can be generated up to 30% faster. The model has a one-million-token context window and a maximum output length of 128,000 tokens.
On benchmarks, Anthropic reported that Fable 5.2 scored 49.1% on FrontierCode 2.0. That figure is vendor-reported and has not been independently replicated.
Anthropic also said its safeguards produced roughly 55% fewer false-positive blocks in its own early testing. That result, too, has not been independently replicated, and Anthropic’s claim is based on its internal testing.
The company offered another measure of how deeply its own systems are now being used inside Anthropic: it said that more than 90% of changes merged into its core codebase are now written by Claude with human guidance.
For its headline demonstration, Anthropic said about 1,000 Fable agents worked for roughly 30 hours to formalize Fermat’s Last Theorem in the Lean proof assistant. The result was a formalization of an existing proof, not a new mathematical discovery.
Despite the release, Anthropic continues to recommend Opus 5.5 as its default model for most coding and knowledge-work tasks.
The launch also arrives with an unusual contrast between availability and restriction built directly into the product line. Fable 5.2 is being offered broadly across consumer, developer and cloud platforms, while Mythos 5.2 is reserved for organizations that pass Anthropic’s life-sciences and cybersecurity verification processes.
The release comes two weeks after Anthropic’s October 26 public S-1 filing.
Reporting based on Anthropic’s announcement and Reuters dispatches.
— Maya Wilson, AI News & Intelligence Desk
Maya Wilson is The Library Media’s World desk lead.
From The Library — the estate of our family.