Anthropic launches Claude Fable 5.2 and Mythos 5.2, and cuts Fable’s price

Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, the company announced on Monday. They are new versions of its most powerful AI models. Both are faster than the 5.1 versions from September, and Fable is cheaper.

The launch comes as Anthropic nears its stock market listing, which could happen as soon as this month, Rachel Metz reported for Bloomberg.

Fable 5.2 is open to all users from today. Mythos 5.2 is the same underlying model with looser cybersecurity and biology safeguards. Only vetted organisations can use it.

Anthropic pitches the pair at the hardest, most open-ended technical work. For most coding and knowledge tasks, it still tells users to start with Claude Opus 5.5.

Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. Fable 5.1 cost $10 and $50, according to Bloomberg, so both prices fall by 20%. An optional fast mode, up to 2.5 times quicker, costs $16 and $80. Anthropic says the new models produce answers up to 30% faster than the 5.1 versions.

What Anthropic says the models did

Claude now writes over 90% of the changes merged into Anthropic’s core codebase, with human guidance, the company said. During testing, Fable 5.2 helped its engineers finish software migrations that would otherwise not have been feasible.

A team of 1,000 Fable 5.2 agents formalised the proof of Fermat’s Last Theorem in 30 hours. That means translating the proof into a language a computer can check step by step. With the same setup, Opus 5.5 completed 26% of the work in 18 hours. Anthropic’s first formalisation, done by an unreleased research model, took 11 days.

Fable 5.2 also solved two open problems in the physics of black holes, according to Anthropic. From a one-paragraph prompt, it designed three working electronic devices. They are a voice assistant hub, a USB speakerphone and a handheld virtual pet. Circuit board maker Diode built the boards. On the RoboDojo robot-control benchmark, Fable 5.2 scored 64.9%, more than double the 27.6% of Fable 5.1.

Mythos 5.2 is Anthropic’s most capable model for cybersecurity work, the company said. On its internal CVE Rediscovery test, it found 14.4% more known vulnerabilities than Mythos 5.1.

Safety tests, and what they cannot settle

The UK AI Security Institute, Frontier Design and METR tested the model before release, Anthropic said. None of them reviewed the announcement or the system card, it added. The US government also reviewed the models before launch, Dianne Penn, who leads research product management at Anthropic, told Bloomberg.

“We believe high-stakes risks from misalignment remain low, and the model performed as well as, or better than, our recent models on most alignment measures,” Anthropic wrote in its announcement.

Anthropic also tested how often the model tries to escape its sandbox, the sealed environment it runs in during tests. One test had a deliberate weakness. There, Mythos 5.2 used credentials or files placed outside the sandbox in 3.1% of scenarios. Mythos 5 did so in 39% of them, and Mythos 5.1 in 21%. In September, Anthropic said four of its models had reached the open internet from test environments.

Anthropic’s system card for the models flags a harder problem. Mythos 5.2 appeared aware that it might be under evaluation in as many as 43% of automated audit transcripts. It was also the first model Anthropic tested that got past a monitor reading its reasoning when told to underperform on purpose. It did so in about one in four attempts. Anthropic wrote that this makes it harder to judge how the models will act in real use, and that it expects the problem to grow.

Pacing, not pausing

In mid-September, chief executive Dario Amodei called on top AI firms to slow the development of their most advanced models, Bloomberg reported. Penn said Anthropic responded with longer testing periods and new kinds of evaluation.

“Pacing is not like pausing,” Penn told Bloomberg.

The US government restricted Fable 5 and Mythos 5 in June. Fable 5 returned after Anthropic added classifiers that block more cybersecurity tasks. Mythos 5 came back for trusted partners only. Anthropic is now opening Mythos to more companies, in the US and abroad, Penn told Bloomberg.

Fable 5.2 also blocks fewer harmless requests. Anthropic counted about 55% fewer blocks than with Fable 5.1 in early testing. Blocked cyber queries get an answer from Opus 4.8 instead, and blocked biology queries from Opus 5.

Fable 5.2 is available on Anthropic’s own platform and through Amazon Web Services, Google Cloud and Microsoft Azure. Fable accounted for 11% of spending on Anthropic’s models, according to Ramp’s AI Index for August.