Summary
Anthropic says Claude Fable 5.2 is available across its platforms, while Mythos 5.2 is offered to vetted organisations working in cybersecurity and life sciences. The company reports gains in coding, scientific reasoning and other technical tasks.
Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, describing them as updated models for demanding technical work. The company says Fable 5.2 is available across its platforms, while Mythos 5.2 is restricted to vetted organisations in cybersecurity and life sciences. Anthropic reports improvements in coding, reasoning and vision, alongside lower prices and faster generation than the previous versions.
The two products use the same underlying model, according to Anthropic. Fable includes safeguards intended to block high-risk cybersecurity tasks and most research-biology tasks; Mythos is made available through verification programmes for organisations working in those areas.
Contents
Benchmark results and technical work
Anthropic’s published evaluations show Fable 5.2 scoring 49.1% on FrontierCode 2.0 at Xhigh effort. The company describes FrontierCode as a test of whether proposed code changes would be accepted by maintainers of open-source repositories. In its results table, Fable 5.1 scored 32.4% and Opus 5.5 scored 45.5%; the table reports each model’s highest available effort unless otherwise noted, so the scores are not all from identical effort settings.
The company also reports a 64.9% score for Fable 5.2 on RoboDojo Sim, a benchmark using a simulated two-arm robot controlled from camera images. Anthropic says this result came from its internal runs at the model’s highest available effort. These tests measure performance on particular tasks and setups, rather than a single general measure of model capability.
Anthropic describes several engineering demonstrations, including a team of agents that produced an operating system able to boot on a laptop, and three working electronic devices designed with circuit boards, 3D-printed enclosures and software. The company says the device boards were manufactured with Diode and that the devices worked without electrical modifications. These are company-reported projects, not benchmark scores.
Research demonstrations
One example concerns formalising Fermat’s Last Theorem in Lean, a programming language that allows a computer to check a mathematical proof step by step. Anthropic says about 1,000 Fable 5.2 agents completed the formalisation in 30 hours, using minimal human intervention. The work encodes a proof for machine checking; it is not a new mathematical theorem. In the company’s comparison, a similarly sized team of Opus 5.5 agents had completed 26% of the formalisation after 18 hours. Anthropic reports one run per model, which makes the result a specific demonstration rather than a repeated-trial estimate.
For protein modelling, Anthropic says Mythos 5.2 used existing open-source folding models and known capsid symmetry to combine predictions for overlapping groups of protein chains. It accurately modelled 28 of the 34 capsids it attempted, using the company’s stated criterion of a structural difference within 3 Å of the experimental structure. This is a computational result about predicted structures, not a test of gene-therapy performance.
Availability, pricing and safeguards
Fable 5.2 is available on Claude’s platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. Developers can access it through the Claude Platform. Anthropic lists prices of $8 per million input tokens and $40 per million output tokens, and says the models generate outputs up to 30% faster than Fable 5.1 and Mythos 5.1. An optional fast mode costs $16 per million input tokens and $80 per million output tokens; Anthropic says it can increase token-generation speed by up to 2.5 times.
Mythos 5.2 access is available through Anthropic’s Cyber Verification Program and Life Sciences Verification Program. Claude Enterprise customers can also use Claude Security, which the company says now includes Mythos 5.2. Anthropic says Fable’s cyber and biology safeguards produce fewer unnecessary blocks than in Fable 5.1, while continuing to block specified high-risk activities. The company reports that its prerelease testing found the underlying model performed as well as or better than recent models on most alignment measures, while noting that behavioural evaluations cannot capture every possible failure before deployment.