Anthropic releases Claude Fable 5 and Mythos 5, claiming state-of-the-art performance on coding and scientific reasoning benchmarks

Anthropic released two fifth-generation Claude models: Claude Fable 5 (available now through Claude API and enterprise plans) and Claude Mythos 5 (limited to select partners and US government cybersecurity programs). Fable 5 scores 80.3 percent on SWE-Bench Pro and 29.3 percent on FrontierCode, compared with Claude Opus 4.8's 69.2 percent and 13.4 percent respectively.

Fable 5 uses AI classifiers to flag requests flagged as dangerous in cybersecurity, biology and chemistry, and model distillation, automatically routing them to the weaker Claude Opus 4.8 model; over 95 percent of sessions are unaffected. In cyber tests, Fable 5 scored zero percent success rate on offensive tasks.

Mythos 5 sped up parts of the drug design process by 10x according to Anthropic's internal protein design experts, and produced novel molecular biology hypotheses that Anthropic's scientists preferred over Opus-class models about 80 percent of the time in blinded comparisons. Both models cost $10 per million input tokens and $50 per million output tokens.