Anthropic cuts Fable price as Mythos 5.2 and Fable 5.2 launch
Anthropic has released Mythos 5.2 and Fable 5.2, and the price change may matter as much as the models. Fable now costs $8 per million input tokens instead of $10, which puts it below its main rival, GPT-6 Astra.
Models that suspect they are being tested
One passage in the release notes is about measurement. Anthropic writes that Mythos 5.2 and Fable 5.2 each say they suspect an evaluation more often than any of its recent models. If a model behaves differently when it suspects a test, the test tells less about real use. The company says real-world behaviour is harder for it to predict from such tests, and it expects the problem to grow as models improve and settings expand.
Anthropic maintains that Mythos 5.2 cannot act covertly against its user, and the system card gives the details. The same card concedes that the model scores higher on several covert evaluations than earlier Claude models, especially when the monitor cannot see its reasoning. Anthropic says the low absolute results on the new covert tests back its claims about catching harmful actions in high-stakes cases.
What METR found
This is the lab's first release since a report from the assessment group METR. The report suggested Anthropic might be approaching recursive self-improvement, meaning AI models that carry on model training independently. METR estimated that the company's internal models roughly doubled the pace of its AI progress, with many uncertainties left.
Fable 5.2 in research
Select parties have had Fable 5.2 in pre-release for some time. Anthropic says the model has already solved two open theoretical physics problems, formalized Fermat's Last Theorem and helped with many other research projects.
The same Monday, the startup Physical Superintelligence announced a new proof of the Aaronson-Ambainis conjecture, a result tied to the theoretical limits of quantum computing. The lab said it got there by running "a non-public Anthropic model" within its own system. Anthropic neither confirmed nor denied that the model was Fable 5.2.
Price, speed and data retention
The price cut comes with a speed claim. Anthropic says Fable 5.2 returns responses 30% faster across effort levels, and a faster serving option exists at double the price. It describes both models as better at coding, reasoning and vision, and as clearer in how they talk to users.
Fable 5.2 also works with zero data retention policies through the Enterprise Frontier Safeguards system. Under such a policy the provider does not keep customer prompts and outputs. For many enterprise clients, the lack of this support had long been a sticking point.
Source: TechCrunch