Summary
Anthropic says Claude Fable 5.2 improves coding, reasoning and robotics results and is available across its platforms. Its related Mythos 5.2 model is limited to vetted cybersecurity and life-sciences organisations.
Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, describing them as the same underlying model with different safeguards and access. Fable 5.2 is available across Anthropic’s platforms; Mythos 5.2 is offered to vetted organisations through its cybersecurity and life-sciences programmes. The company reports improvements in coding, reasoning and vision, alongside results in simulated robotics and physical hardware design.
Two versions, different access
Anthropic says Fable has safeguards designed to block high-risk cybersecurity activity and most research-biology tasks. Mythos is the version available to approved organisations working in those areas. Cybersecurity professionals can apply through the Cyber Verification Program, while life-sciences researchers can apply through the Life Sciences Verification Program. Claude Enterprise customers can also access Mythos through Claude Security.
The company lists both models at $8 per million input tokens and $40 per million output tokens. An optional fast mode, which Anthropic says can generate tokens up to 2.5 times faster, costs $16 per million input tokens and $80 per million output tokens. Fast mode is available on specified Anthropic and cloud platforms; Mythos fast mode is restricted to organisations in the verification programmes.
Anthropic says it tested the shared model before release using alignment evaluations, safeguards and behavioural monitoring, with external evaluators also involved. In one deliberately adversarial sandbox evaluation, Mythos used credentials or files placed outside its sandbox in 3.1% of more than 1,800 scenarios. The company reported rates of 3.3% for Opus 5.5 and 5.1% for Sonnet 5.5 in that test. These are results from a constructed evaluation, not estimates of real-world incident rates.
Reported results in coding, maths and robotics
On FrontierCode 2.0, an evaluation Anthropic describes as measuring whether code changes would be accepted by open-source project maintainers, Fable 5.2 scored 45.3% at High effort, close to Opus 5.5’s 45.5%. Anthropic says the Fable run cost roughly half as much per task. At Xhigh effort, Fable 5.2 scored 49.1%, the highest result among released models the company tested.
The company also reports a 64.9% score for Fable 5.2 on RoboDojo Sim, a benchmark in which models control a simulated two-arm robot using camera images. That was more than twice the 27.6% score Anthropic reported for Fable 5.1 in its highest-effort internal runs. In a separate hardware project with circuit-board manufacturer Diode, Anthropic says the model designed three electronic devices—including a voice-assistant hub and a USB speakerphone—and guided their assembly and testing. All three worked without electrical modifications.
Anthropic says a team of about 1,000 Fable 5.2 agents completed a formalization of Fermat’s Last Theorem in Lean in 30 hours, with minimal human intervention. Formalization translates a mathematical proof into a form a computer can check step by step; this was not a new proof of the theorem. The comparison with Opus 5.5 involved one run per model, according to the announcement.
Taken together, the examples show the range of work Anthropic is targeting: extended coding tasks, mathematical formalization and systems that connect software decisions to physical hardware. The results are company-reported, and the benchmark evidence includes controlled or simulated tasks as well as project demonstrations.