Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, the identical mannequin with totally different safeguards for coding, data work, and scientific analysis. Fable is usually accessible now, whereas Mythos is proscribed to vetted cyber defenders and life scientists. Builders can name claude-fable-5-1 through the Claude API and entry Fable by means of Anthropic’s merchandise, AWS, Google Cloud, and Microsoft Azure.
We’re introducing Claude Fable 5.1 and Claude Mythos 5.1.
They’re the world’s most superior fashions for coding and data work. pic.twitter.com/8P9PSrWPi3
— Claude (@claudeai) September 1, 2026
Cache reads now price 75% much less at $0.25 per million tokens. Anthropic estimates typical prices will fall 25%, with financial savings on extremely agentic work reaching about 45%. Different charges stay $10 per million enter tokens and $50 per million output tokens. Fable defaults to Excessive effort in Claude Code and Medium effort in Claude Cowork and Claude.
Anthropic additionally launched Enterprise Frontier Safeguards, which hold buyer knowledge inside infrastructure managed by the shopper whereas retaining misuse protections. It’s going to roll out to enterprises in phases later this fall. Till then, eligible prospects can use Fable with zero knowledge retention. Anthropic studies 60% fewer false-positive cyber interventions per Claude Code session and 85% fewer biology interventions on benign elementary questions than its earlier safeguards.
In addition to being able to a lot larger efficiency than Fable 5, it will possibly additionally obtain comparable or higher outcomes at a a lot decrease price when set to decrease effort ranges. pic.twitter.com/xc7sBvEx5Y
— Claude (@claudeai) September 1, 2026
Anthropic studies Fable scored 52.6% on Terminal-Bench-Science 0.1 at most effort, versus 24.7% for Fable 5, and 73.4% on CursorBench 3.2.0. Mythos reached 60.9% on Terminal-Bench 4.0, in contrast with 55.8% for Fable. Manufacturing safeguards had been lively and will decrease some scores.
In analysis exams, Mythos designed protein binders despatched for out of doors validation, with Anthropic reporting successful fee close to 50% throughout 12 targets. Fable educated a neural community to supply a Venus elevation map at two-to-three-kilometer decision, up from 10 to twenty kilometers, whereas Mythos wrote GPU kernels for seven open-source biology fashions with speedups as much as 2.5 occasions. The map can be launched beneath Inventive Commons, and Anthropic plans to open-source the biology optimizations.
Anthropic examined each fashions for chemical and organic threat, cybersecurity, agentic security, alignment, immediate injection, and reward hacking. Mythos entry runs by means of the Cyber Verification and Life Sciences Verification applications and powers Claude Safety. New-model outputs carry an invisible numerical watermark for EU AI Act transparency, backed by a private-preview detection API for eligible teams.
