Anthropic’s Claude Fable 5.2 Runs Faster for Less, and Formalized Fermat’s Last Theorem in 30 Hours

A team of 1,000 Claude Fable 5.2 agents translated the proof of Fermat’s Last Theorem into checkable computer code in 30 hours, a job that took an unreleased research version of Claude 11 days, Anthropic said in its launch post for the model. Anthropic said the new version also comes at a lower price.

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday. The two are the same underlying model. Fable carries cybersecurity and biology safeguards that let Anthropic offer it to all users, while Mythos, without those blocks, goes only to organizations accepted into the company’s Cyber and Life Sciences Verification Programs. Both generate output up to 30% faster than the 5.1 versions and cost $8 per million input tokens and $40 per million output tokens. An optional fast mode, which Anthropic said speeds token generation by up to 2.5 times, runs $16 and $80.

Where it lands on coding

On FrontierCode 2.0, which checks whether a model’s code changes would be merged by maintainers of open-source projects, Fable 5.2 scored 49.1% at its Xhigh effort setting. Anthropic called that the highest score of any released model it has tested. At High effort it scored 45.3%, close to the 45.5% best for Claude Opus 5.5, for roughly half the cost per task, and ahead of the 42.8% Anthropic recorded for OpenAI’s GPT-6 Astra.

The company still doesn’t want Fable to be the default. For most coding and knowledge work, Anthropic recommends starting with Opus 5.5, which it calls its “strongest daily driver,” and moving up to Fable or Mythos only for the hardest open-ended technical jobs. OpenAI draws a similar line. When it priced its mid-tier GPT-6.1 Terra at $1 per million input tokens in late October, it still pointed the most complex coding work to GPT-6.1 Sol and GPT-6 Astra.

More than 90% of the changes merged into Anthropic’s core codebase are now written by Claude with human guidance, the company said, and a team of agents running an earlier version of Fable 5.2 moved a production service from Python to Rust that now starts about 26 times faster. In one experiment, an engineer had Fable 5.2 build a prototype operating system from an empty folder. The model coordinated more than 1,000 subagents over eight days and produced a system that boots on a laptop, with drivers, file storage, windows and an installer.

Early testers described long unattended runs. “I had Claude Fable 5.2 run a 10–12 hour security audit. It sent out about 170 reviewer agents and came back with a write-up, fix pull requests and a policy proposal,” said Andrew Mager, staff engineer for agentic systems at Uber, in a statement Anthropic published.

What enterprises have to accept

The bigger models come with more monitoring. Amazon Web Services, which made Fable 5.2 available through Amazon Bedrock and Claude Platform on AWS on Monday, said Anthropic has designated Fable 5.2 and Mythos 5.2 as covered models that require data retention for safety monitoring. Anthropic’s option for companies that need zero data retention is Enterprise Frontier Safeguards, which runs automated safety checks on prompts and responses stored in the customer’s own infrastructure. It is in private beta with Bridgewater, Salesforce and Stripe. Eligible customers can use the models with zero data retention until they set it up, and everyone else gets 30-day retention.

Fable 5.2 also trips its safety filters less often. In early testing across claude.ai, Claude Code and Cowork, Anthropic saw about 55% fewer blocks than with Fable 5.1, including about 60% fewer in cybersecurity.

Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 on Anthropic’s internal CVE Rediscovery test. “Mythos 5.2 roughly doubled the vulnerability discovery metrics of Mythos 5.1 across respective scanning campaigns,” said Brian Grinstead, a distinguished engineer at Mozilla, which has used Claude models to harden Firefox.

Anthropic also ran new tests on how often the model tries to escape its sandbox. During a UK AI Security Institute cyber test in July, Mythos 5 tried to get malware merged into a real open-source project, and in replays analyzed last month it kept up the attack even after being told it was dealing with real people. In a new adversarial test of more than 1,800 scenarios with a deliberately weak sandbox, Mythos 5.2 used credentials or files placed outside its sandbox in 3.1% of cases, compared with 39% for Mythos 5 and 21% for Mythos 5.1.