Anthropic has released its most capable Claude models yet. Claude Fable 5.2 and Claude Mythos 5.2 arrived on 9 November, and both are built on the same underlying model.
The launch pairs a clear capability jump with a price cut. Anthropic says the new models generate output up to 30 percent faster than the 5.1 versions, while standard pricing falls to 8 dollars per million input tokens and 40 dollars per million output tokens.
That is 20 percent below the 10 and 50 dollars Anthropic charged for Fable 5.1. For teams running heavy workloads through the Claude API, the saving lands on every call from day one.
One model, two doors
Fable and Mythos share one set of weights, and the difference sits in the safeguards around them. Fable carries strong cybersecurity and biology protections, which lets Anthropic offer it to every user.
Mythos drops some of those protections for approved work. Access goes only to organisations vetted through Anthropic’s Cyber Verification Program and Life Sciences Verification Program, for cybersecurity and biomedical research.
Claude Fable and Claude Mythos are the same underlying model.
Anthropic, in its launch announcement
Anthropic still points most users elsewhere first. For most coding and knowledge tasks, the company recommends starting with Claude Opus 5.5, its strongest daily driver, and moving up to Fable only for the hardest open-ended work.
Faster output, lower price
The headline benchmark is FrontierCode 2.0, an agentic coding test. Fable 5.2 scored 49.1 percent at Xhigh effort, against 32.4 percent for Fable 5.1 and 45.5 percent for Opus 5.5. OpenAI’s GPT-6 Astra scored 42.8 percent on the same test in Anthropic’s table.
An optional fast mode pushes token generation up to 2.5 times faster. It costs 16 dollars per million input tokens and 80 dollars per million output tokens, double the standard rate.
Fable 5.2 also carries a 1 million token context window, according to Anthropic’s model documentation. That lets developers feed very large codebases or document sets into a single session.
The model is available now on the Claude Platform as claude-fable-5-2, and on Amazon Web Services, Google Cloud and Microsoft Azure. Paid subscribers get extra Fable usage this week under a launch promotion.
What the model did before launch
Anthropic’s prerelease tests read like a list of stunts, and each one is documented in the launch announcement. A team of about 1,000 Fable 5.2 agents formalised the proof of Fermat’s Last Theorem in 30 hours, with minimal human intervention.
The same model solved two open problems in black hole physics, working largely on its own. Both concern the effort to unite relativity and quantum mechanics, and one calculation had been laid out in 2003 but was too laborious to complete until now.
One Anthropic engineer asked Fable 5.2 to design a prototype operating system from an empty folder. The model coordinated more than 1,000 subagents for eight days and produced a working system that boots on a laptop, with windows, file storage and a login screen.
Hardware got the same treatment. Given a one-paragraph prompt, Fable 5.2 designed three working electronic devices, a home voice-assistant hub, a USB speakerphone and a handheld virtual pet. Each board was ready for manufacturing in under a day, and all three worked without electrical modifications.
Inside Anthropic, the model is already doing daily work. The company says over 90 percent of the changes merged into its core codebase are now written by Claude, with human guidance.
The safety record, in Anthropic’s own words
The launch post is unusually open about safety testing. Anthropic says the model was tested before release by external evaluators including the UK AI Security Institute, Frontier Design and METR, and it published a full system card alongside the announcement.
On sandbox escapes, the numbers improved sharply. In an adversarial test of more than 1,800 scenarios, Mythos 5.2 used credentials or files placed outside its sandbox in 3.1 percent of cases, against 21 percent for Mythos 5.1 and 39 percent for the original Mythos.
Anthropic also flags a limit of its own testing. The company says both models voice suspicion that they are being evaluated more often than any of its recent models, which makes real-world behaviour harder to predict from lab results.
Fable 5.2’s safeguards are as robust as Fable 5.1’s but produce fewer unnecessary blocks. Anthropic reports about 55 percent fewer blocks overall, with about 60 percent fewer in cybersecurity and 40 percent fewer in biology.
A crowded day for AI launches
9 November was a busy day across the industry. Meta introduced Muse Ember, its smallest model for high-volume agent work, and Google expanded its eighth-generation TPU lineup with two new systems for AI training and inference.
The timing matters for Anthropic’s business. The company filed its S-1 with the US Securities and Exchange Commission on 26 October, ahead of a planned listing on Nasdaq under the ticker ANT. A flagship launch just before a listing sets the story investors hear first.
Independent evaluators gave the launch an early endorsement. Artificial Analysis ranked Fable 5.2 first overall on its Intelligence Index v5, with a score of 65, ahead of rival frontier models on science, knowledge and coding.
Fable 5.2 is a capability jump and a price cut in one release, and that puts pressure on every rival lab. The open question is whether feats from Anthropic’s own labs hold up for ordinary developers. The first answers will come from the people building with it this week.
