Anthropic ships Fable 5.2 and Mythos 5.2 amid scrutiny of self-improving models
Seven Lights
Anthropic released new versions of its two most capable artificial-intelligence systems on Monday, making Claude Fable 5.2 and Claude Mythos 5.2 available with improved coding, reasoning and vision performance.
Both models generate outputs 30 per cent faster than the 5.1 versions and “communicate more clearly,” Anthropic said. Pricing fell below the predecessors, to eight dollars per million input tokens and 40 dollars per million output tokens. Fable 5.2 launched for all Claude users, with what the company called strong cybersecurity and biology safeguards. Mythos 5.2, the more capable of the pair, is available only to verified organizations, in areas such as cybersecurity.
The release carries guardrails that block risky cybersecurity activities, including penetration testing and the generation of exploits, and advanced biology capabilities. Eligible Claude subscribers will receive one to three additional usage resets for Fable this week, based on the length of their subscriptions.
The launch arrives under unusual scrutiny of the industry’s own momentum, a question known as recursive self-improvement, or the ability of AI models to carry on model training independently. The evaluation group METR found that Anthropic’s internal models were speeding up the company’s AI progress by roughly two times, though many uncertainties remain. Anthropic’s release notes acknowledged continuing challenges in testing model capabilities. “Claude Mythos 5.2 and Fable 5.2 each voice suspicion that they are being evaluated more often than any of our recent models, which challenges our ability to assess how they will act in real-world settings,” the company said, adding that it expects the difficulty to grow as capabilities increase. Anthropic maintains that Mythos 5.2 does not have the ability to act covertly against its users, a finding detailed in the model’s system card.
The release follows recent frontier launches by OpenAI, Google and Meta and sharpens competition for enterprise contracts. Anthropic said the restricted availability of Mythos reflects a wider effort by leading labs to stage access to their most capable systems.