Claude Fable 5.1

Anthropic's second public Mythos-class release, succeeding Claude Fable 5 as the broadly available counterpart to the restricted Claude Mythos 5.1. #1 on the AA Intelligence Index v4.3 at 53 (57 on v4.2; 66 on v4.1.1; Opus 5 scores 51). Headline gains are agentic and scientific: Terminal-Bench 4.0 55.8 (Fable 5 42.0, Opus 5 52.3), Terminal-Bench-Science 0.1 52.6 — more than double Fable 5's 24.7 — GDPval-AA v2 Elo 1853 (1723/1824), OSWorld 2.0 strict 41.7, HLE without tools 60.9, AutomationBench 31.4 (17.1/26.9). Anthropic showcases research results: ~50% protein-binder hit rate across 12 targets, GPU-kernel optimizations up to 2.5×, and a CC-licensed 2–3 km Venus elevation map.

1M-token context, 128K output, always-on adaptive reasoning with low/medium/high/xhigh/max effort (high is the Claude Code default, where it is now the default Fable model). Priced $10/$50 per million tokens with cache reads cut 75% to $0.25/M — about 25% cheaper on typical workloads and up to 45% on agentic ones. Safeguards are refined rather than lifted: blocked cybersecurity tasks fall back to Opus 4.8 and blocked biology tasks to Opus 5, with 60% (cyber) and 85% (bio) fewer false positives than Fable 5's launch safeguards, plus anti-distillation mechanisms on new API accounts. Anthropic assesses it below the next RSP capability tier. API ID claude-fable-5-1; knowledge cutoff June 2026.