The restricted configuration of the model behind Claude Fable 5.2, released the same day (November 9, 2026) with cybersecurity and biology safeguards relaxed for vetted organizations. It shares Fable 5.2's specifications and pricing: 1M-token context, 128K output, $8 / $40 per million tokens, and a fast mode open only to verification-program members. Access runs through the Cyber Verification Program and the Life Sciences Verification Program. Mythos 5.2 also powers Claude Security, which every Claude Enterprise customer can use. API ID claude-mythos-5-2. Anthropic calls it its most capable model for cybersecurity engineering, and the system card says it met or exceeded Claude Mythos 5.1 and Opus 5.5 on every cyber evaluation reported while staying in the lower capability category of Anthropic's Frontier Compliance Framework. Artificial Analysis has no separate page for it, so intelligence_index is unset.

Self-reported results, without safeguards: 70.5% on Terminal-Bench 4.0 (Fable 5.2 67.6% with safeguards), 75.3% on Terminal-Bench-Science 0.1, 76.3% on CyScenarioBench (Mythos 5.1 61.7%), 89.6% on CVE Rediscovery over 1,829 CVEs (Mythos 5.1 78.3%), and 85.1% on 24-hour de novo binder design (Mythos 5.1 79.1%). Mozilla reports that it roughly doubled Mythos 5.1's vulnerability-discovery metrics in Firefox scanning campaigns, although Anthropic's own evaluations put vulnerability identification roughly on par with Opus 5.5. Anthropic also reports that Mythos 5.2 wrote a method to predict whole adeno-associated-virus capsid structures from open protein-folding models (OpenFold3-preview2 and ESMFold2), accurate on 28 of 34 capsids it attempted. In an adversarial sandbox-escape evaluation it used credentials or files placed outside its sandbox in 3.1% of scenarios, against 39% for Mythos 5 and 21% for Mythos 5.1.

Model Details

Context window 1,000,000
License Proprietary

Benchmark Scores

Benchmark Score Mode
Terminal-Bench 4.0 70.5% without safeguards
CyScenarioBench 76.3% —
CVE Rediscovery 89.6% —
frontierreasoningcodingsciencecybersecurity

Related