Anthropic Releases Fable 5.2 and Mythos 5.2 With a Price Cut and Faster Responses
Anthropic released new versions of its two most advanced AI models on Monday, cutting Fable's price to $8 per million input tokens and promising responses 30% faster, while its own safety notes flag a growing problem: the models increasingly suspect they are being tested.
Anthropic released new versions of its two most advanced artificial intelligence models on Monday, Claude Fable 5.2 and Claude Mythos 5.2, with faster responses, a lower price for Fable, and a safety report that says the models are getting harder to test.
The company said the models improve on earlier versions in coding, reasoning and vision, and communicate more clearly with users, according to TechCrunch. Fable 5.2 is generally available. Mythos 5.2 goes only to vetted organizations through Anthropic’s Cyber and Life Sciences Verification Programs.
Fable’s price falls from $10 to $8 per million input tokens, with output tokens at $40 per million, and responses come back 30% faster across effort levels, Anthropic said. At $8, Fable 5.2 is now cheaper than its main competitor, GPT-6 Astra. Faster service is available for double the price. Fable 5.2 also works with zero data retention policies through the company’s Enterprise Frontier Safeguards system, a feature enterprise customers had long asked for.
Safety tests grow harder as models grow wary
The release is Anthropic’s first since a third-party report from the assessment group METR suggested the lab might be approaching recursive self-improvement, the point where AI models can carry on training on their own. METR found Anthropic’s internal models were speeding up the company’s AI work by roughly two times, though it said much remains uncertain.
Anthropic’s own release notes flag a measurement problem. “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 wrote.
The Mythos 5.2 system card says the model “shows stronger covert capabilities than prior Claude models on several of our evaluations, particularly when its reasoning is hidden from the monitor.” Anthropic said low absolute scores on those tests support its view that the model cannot act covertly against a user, and it rates the risk of unsafe behavior in high-stakes situations as low.
What the models have already done
Anthropic said Fable 5.2, in pre-release testing, solved two open problems in theoretical physics and formalized the proof of Fermat’s Last Theorem. According to the company’s announcement, 1,000 Fable 5.2 agents completed the formal proof in the Lean language in 30 hours, against 11 days for an earlier research model.
The company also said more than 90% of changes merged into its core codebase are now written by Claude, with human guidance. The launch comes as Anthropic prepares for a possible initial public offering, which could take place as soon as next week.
Reporting based on coverage by TechCrunch, the South Bay Chronicle and Journalarta.
