Anthropic introduces Claude Fable 5.2 and Mythos 5.2 for technical work
Anthropic introduced the Claude Fable 5.2 and Claude Mythos 5.2 models. The first is available through major cloud platforms, while the second has restricted access for verified organizations.
Claude Fable 5.2 is Anthropic’s new publicly available model designed for demanding technical work. On November 9, the company also introduced Claude Mythos 5.2, but restricted access to verified organizations participating in its cybersecurity and biological sciences programs.
Fable 5.2 is available directly through Claude Platform, as well as through AWS, Google Cloud, and Microsoft Azure. Anthropic charges $8 per million input tokens and $40 per million output tokens. The optional fast mode costs twice as much.
Claude Fable 5.2 leads the Artificial Analysis index
The independent service Artificial Analysis assigned Claude Fable 5.2 a score of 65 in its Intelligence Index. In its evaluation of 74 models, it ranked the model first.
However, this result should be read in the context of the methodology and model configuration used. A score from a single benchmark index alone does not confirm that the model will be best for every practical use, or that it has a general advantage over competitors.
Anthropic describes its new models as systems for technically demanding tasks. Claims of significantly better programming, scientific reasoning, electronics design, or work on open problems in physics are, however, based primarily on the company’s announcement and preliminary testing so far. This is not independently reproduced confirmation.
Mythos 5.2 has restricted access
While Anthropic provides Claude Fable 5.2 through standard product and cloud channels, Claude Mythos 5.2 is not freely available. Access is intended only for verified organizations in Anthropic’s cybersecurity and biological sciences programs.
The company thereby distinguishes between a publicly deployed model and a system with limited distribution for more sensitive areas. The published information does not indicate expanded access beyond these programs.
METR assessment context
The models’ release comes several days after the publication of an assessment by METR concerning the acceleration of AI research and development at Anthropic. TechCrunch characterized the findings as roughly a twofold acceleration of internal AI progress, while also pointing to continuing uncertainties.
The assessment also appeared in discussions about possible recursive AI self-improvement—meaning a situation in which systems accelerate their own development. However, the available materials do not confirm that Anthropic has achieved full recursive self-improvement. The METR report is relevant context in this debate, not evidence of such a capability.
What will be important to watch
- independent comparisons of Fable 5.2’s performance with competing models,
- external assessments of the security measures and system cards for both models,
- possible changes to the access rules for the Mythos 5.2 model,
- further clarification from METR on the scope, causes, and limits of the acceleration of AI research and development at Anthropic.
Sources
- Anthropic – Confirms the models’ release, availability, access modes, pricing, and the company’s own claims about their capabilities and safety.
- METR – The original source of the independent assessment of the acceleration of AI research and development at Anthropic.
- Artificial Analysis – Reports a score of 65 and first place for Claude Fable 5.2 in its Intelligence Index.
- The Verge – Independently confirms the announcement and the distinction between Fable 5.2’s availability and the restricted access to Mythos 5.2.
- TechCrunch – Provides context on the recent METR assessment and its uncertainties surrounding RSI.
Verified and updated: 11/10/2026 06:20
