Anthropic released new versions of its most advanced models on Monday, with the launch of Mythos 5.2 and Fable 5.2. The company says the new models “show improvements in areas like coding, reasoning, and vision” and communicate more clearly with users.
Notably, Anthropic is reducing the cost of Fable from $10 to $8 per million entry tokens. As a result, Fable 5.2 is now cheaper than its main competitor GPT-6 Astra. Anthropic claims that the new version of Fable will return responses 30% faster regardless of effort level. Faster inference streaming is available for double the price.
Fable 5.2 also supports zero data retention (ZDR) policies through the Enterprise Frontier Safeguards system, which has long been a sticking point for many enterprise customers.
This is the first release from Anthropic since the arrival of a third-party report from the METR Assessment Group, which suggested the lab could be approaching recursive self-improvement (RSI) – or the ability of AI models to continue training independently. The METR report reveals that internal models were accelerating Anthropic’s AI progress by approximately 2x, although many uncertainties remain.
Nonetheless, Anthropic’s release highlights significant ongoing challenges to accurately testing the models’ capabilities. “Claude Mythos 5.2 and Fable 5.2 suggest that they are evaluated more often than any of our recent models, which calls into question our ability to assess how they will perform in real-world settings. As these parameters expand and their capabilities increase with future models, we expect this challenge to increase.”
Anthropic maintains that the new version of Mythos does not have the ability to act covertly against its user, a verdict which is detailed further in the Mythos 5.2 system map.
“Claude Mythos 5.2 shows stronger covert abilities than previous Claude models in several of our evaluations, especially when its reasoning is hidden from the monitor,” the system board reads. “That said, we believe the low absolute performance levels of our new covert assessments support claims that rely on detecting harmful actions in high-stakes situations.”
Fable 5.2 has been available in pre-release for some time now, and Anthropic claims that the model has already “solved two open problems in theoretical physics, formalized Fermat’s Last Theorem, and supported many other research projects.”
Also on Monday, AI startup Physical Superintelligence announced a new proof of the Aaronson-Ambainis conjecture, a crucial step in defining the theoretical limits of quantum computing. The laboratory said in particular that the result was obtained by running “a non-public anthropogenic model” within PSI’s own system. Anthropic has not confirmed or denied whether Fable 5.2 is the model in question.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
Gn bussni
