Anthropic launches Claude Fable 5 with strict safety limits and 90% benchmark score

Anthropic has launched Claude Fable 5, making its most powerful AI model available to the general public for the first time. The model is a version of the previously restricted Mythos, which was limited to a few partners due to cybersecurity concerns. Fable 5 excels in software engineering, knowledge work, and vision, scoring 90% on a complex analytics benchmark by Hex—the first model to do so. It also outperforms competitors in UI design, game coding, and one-shotting full apps. However, Anthropic has placed hard safety limits: in high-risk domains like cybersecurity, biology, chemistry, and distillation, Fable 5 defers to the older Opus 4.8 model. Early data shows at least 95% of sessions run entirely on Fable. Pricing is set at $10 per million input tokens and $50 per million output tokens, double the cost of Opus 4.8, which may deter some enterprises already wary of AI costs.

To address safety concerns, Anthropic implemented a 30-day mandatory data retention policy on all traffic, even for customers with previous zero-retention agreements. The company will not use the data for training but only to defend against novel attacks and identify false positives. This policy could set an industry precedent. Anthropic stress-tested the model with over 1,000 hours of bug bounty testing, finding no universal jailbreaks. The launch comes as Anthropic prepares for public markets and calls for a global brake pedal on frontier AI development, warning of recursive self-improvement risks. While the model is free on Pro, Max, Team, and Enterprise plans through June 22, it will then require usage credits until restored as a standard feature.