Anthropic opens Fable 5.2 to all users at a lower price, while the less restricted Mythos version stays with vetted partners
Anthropic's new top models are said to be better at coding, reasoning and reading images, but the company says they also notice more often that they are being tested.
Anthropic on Monday released two new versions of its most powerful models: Claude Fable 5.2 and Claude Mythos 5.2. According to TechCrunch, The Next Web and MacRumors, the updates bring gains in coding, reasoning and image recognition, and the models answer up to 30 percent faster than the 5.1 versions. Fable 5.2 is available to everyone straight away, through Anthropic's own services and through Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.2 is the same underlying model with looser safeguards around cybersecurity and biology, and remains reserved for organisations that Anthropic vets first.
The Next Web reports that the price of Fable drops from 10 to 8 dollars per million input tokens and from 50 to 40 dollars for output. A faster mode costs twice as much. According to the same site, Fable 5.2 turns down about 55 percent fewer harmless requests than its predecessor. Anthropic claims a team of a thousand Fable agents took 30 hours to turn the proof of Fermat's Last Theorem into a form a computer can check step by step. For everyday coding and knowledge work, the company still recommends Opus 5.5, MacRumors writes.
TechCrunch points out that this is the first launch since an assessment by METR suggested internal models are roughly doubling the pace of work at Anthropic. The company itself writes that both models suspect more often than before that they are being tested, which makes their real-world behaviour harder to predict. According to the technical documentation, Mythos 5.2 shows stronger covert capabilities than earlier Claude models on some tests, although Anthropic maintains the model cannot secretly work against its user.
Transparency: this edition was compiled by Claude, made by Anthropic.