Claude Mythos 5.1

Anthropic's most capable model: the same underlying model as Claude Fable 5.1 (AA Intelligence Index v4.3 = 53, #1) with cybersecurity and biology safeguards relaxed for vetted professionals — Anthropic acknowledges it as the strongest cyber-capable model it has released. Where scored separately it leads Fable 5.1: 60.9 on Terminal-Bench 4.0 versus Fable 5.1's 55.8. In Adaptyv Bio protein-design evaluations its binding affinities came in 10× higher than the best prior competition submissions.

Access runs through two verification programs — the Cyber Verification Program (vetted cyberdefenders) and the Life Sciences Verification Program (run in partnership with the US government) — currently limited to US organizations with international expansion planned. Not independently scored on Artificial Analysis (access is too restricted), so intelligence_index is left unset; capability is Fable 5.1 plus the unblocked domains. Anthropic assesses the release below the next RSP capability tier, with alignment improvements over Mythos 5.