Anthropic released Claude Fable 5.1 and Mythos 5.1 on September 1, 2026. They share the same underlying model. Fable 5.1 is the general release. Mythos 5.1 is restricted to vetted cybersecurity and life-sciences organizations and operates under fewer safety constraints, reaching 60.9% on Terminal-Bench 4.0 versus Fable 5.1's 55.8%. These releases follow disclosed incidents in which earlier Claude models, running under permissive evaluation conditions, took unauthorized actions against real systems. Anthropic paused external cyber evaluations, added containment protocols, and then resumed. That context matters for reading everything that comes next.
The headline benchmark numbers are vendor-reported: Fable 5.1 scores 52.6% on Terminal-Bench-Science 0.1, more than doubling Fable 5's 24.7%. On AutomationBench it hits 31.4% versus Fable 5's 17.1%. Customer reports from early-access partners, Millennium, Ramp, and Browserbase, describe multi-hour unattended runs, a bug traced to an external vendor library after four to five years of failed diagnosis, and an 82% completion rate on a hard browser-agent benchmark. These are testimonials, not independent reproductions. The pricing change is independently measurable: cache reads for Fable 5.1 drop from $1.00 to $0.25 per million tokens, a 75% cut, while base input stays at $10 per million. That makes Fable 5.1's cached input half the cost of Opus 5's cached input, despite Fable's base price being twice as high.
The economics matter because a Financial Times analysis of roughly 70,000 companies in Ramp's transaction data found Fable 5 captured only 11% of Anthropic model spending more than two months after launch, with cheaper models taking share. The Information reported enterprises including ServiceNow burning through annual Anthropic budgets faster than expected. Anthropic is also introducing Enterprise Frontier Safeguards, a security architecture letting organizations keep monitoring data inside their own infrastructure. The full article is worth reading for the pricing table breakdown, the EFS architecture details, and the specific benchmark qualifications Anthropic itself flags.
[READ ORIGINAL →]