Claude Mythos 5

Anthropic's most capable model — the Mythos-class tier above Opus, and the same underlying model as Claude Fable 5 (AA Intelligence Index v4.1 = 60) but with safeguards lifted in some domains (cybersecurity entirely, vs Fable 5's fallback to Opus 4.8). Released 2026-06-09 alongside Fable 5. Not independently scored on Artificial Analysis — access is too restricted for evaluation — so intelligence_index is left unset; capability is that of Fable 5 plus the unblocked domains.

Highly restricted availability: initially only Project Glasswing cybersecurity partners and critical-infrastructure providers, with a planned biology trusted-access program for a small number of life-science researchers and a broader cybersecurity program under development in consultation with the US government — not generally available. Anthropic highlights scientific gains: protein design (~10× faster drug design), molecular-biology hypothesis generation (preferred ~80% over Opus-class models), and genomics models that outperform published Science-journal results with ~100× fewer parameters. $10/M input, $50/M output (same as Fable 5).