Anthropic has introduced Claude Fable 5.2 and Claude Mythos 5.2. They are the same model with two access options: Fable 5.2 has enhanced safeguards for cybersecurity and biology and is available to everyone, while Mythos 5.2 is only available to verified organizations through the company’s cybersecurity and life sciences programs, or through Claude Security for enterprises.

The key details from the announcement:

🔹 Price: $8 per million input tokens and $40 per million output tokens. There’s an optional fast mode, up to 2.5 times faster, for $16 and $80.

🔹 Speed: generates results up to 30% faster than version 5.1.

🔹 Availability: AWS, Google Cloud, and Microsoft Azure.

🔹 Fewer refusals: Fable 5.2 rejects 55% fewer queries than 5.1, with equally robust safeguards, according to the company.

🔹 Code: Anthropic says that Claude now writes more than 90% of the changes it integrates into its main codebase, with human oversight.

🔹 Mathematics: 1,000 Fable 5.2 agents formalized Fermat’s Last Theorem in Lean in about 30 hours.

🔹 External evaluation: it was evaluated by, among others, the UK’s AI Security Institute and METR.

In the performance tests published by the company, Fable 5.2 outperforms Opus 5.5 in almost every measured capability. On FrontierCode 2.0, which measures whether open-source project maintainers would accept the model’s changes, it scores 49.1% with very high effort, compared with 45.5% for Opus 5.5 and 42.8% for GPT-6 Astra. Anthropic adds that it deliberately kept its AI R&D capabilities around the level of Opus 5.5.

For everyday use, the company still recommends starting with Opus 5.5 and switching to Fable or Mythos for more ambitious technical work. Fable 5.2 can also act as an adviser to smaller models or as an orchestrator that delegates tasks to them.

What I find most relevant is the tiered-access model: the same technology, but with the most sensitive capabilities reserved for verified organizations. It’s a way to manage risk that we’ll probably see more labs adopt as models become more capable in areas such as cybersecurity.

Full analysis at territorioblockchain.com

#InteligenciaArtificial #Anthropic #Cybersecurity