Anthropic shipped one model in two skins on September 1. Fable 5.1 is the frontier model everyone gets; Mythos 5.1 is the same weights with looser safeguards, reserved for vetted cybersecurity and life-science institutions. Splitting a release by trust tier, not capability, is the actual news here.
The numbers back it up: Terminal-Bench-Science jumps from 24.7% to 52.6%, ARC-AGI-2 hits 90%, and the Artificial Analysis index puts it at 66 — above Opus 5’s 63. Party trick included: it cracked Sir Thomas Urquhart’s 1653 cipher, unsolved for 373 years, in 44 minutes and 176K tokens.
The price cut beats the benchmarks
Cache reads drop 75% to $0.25 per million tokens. Typical agentic workloads get 25–45% cheaper. Smarter model, half the bill — that’s the combination that moves production traffic.
API access
Available now as claude-fable-5-1 on the Claude API, plus AWS, Google Cloud, and Azure, with a 1M-token context window and 128K output — built for long-running agents that hammer the same cached context. Mythos 5.1 requires institutional review. Enterprises get customer-managed storage, zero-retention options, and EU AI Act watermarking.
You Might Also Like
- Deepseek v4 Flash 0731 Hits 82 7 on Terminal Bench Chasing Opus 4 8 at 0 14 m Tokens
- Qwen3 8 max Goes ga Alibabas 2 4t Model Beats Claude Fable 5 on Terminal Bench Weights Open Next Week
- Ornith 1 5 Ties Claude Opus 4 8 on Terminal Bench Deepreinforce Trained it on Tasks it Wrote Itself
- Arc agi 3 Turns ai Testing Into a Video Game and Every Frontier Model is Losing
- Kimi k2 6 Beats gpt 5 4 and Claude Opus 4 6 on swe Bench pro
Top AI Product 
Leave a comment