Anthropic has announced the launch of its newest AI models, Fable 5.1 and Mythos 5.1, addressing customer feedback regarding pricing, data retention, and overzealous safeguards. The company claims that Claude Fable 5.1 delivers stronger performance than Fable 5 while reducing costs by approximately 25 percent typically, and up to 45 percent for complex agentic tasks, primarily due to reduced pricing on cached data.
Performance and Early Feedback
Early impressions from industry leaders highlight significant improvements. Every CEO Dan Shipper described it as the strongest coding model they have used, noting its speed, token efficiency, and natural communication style. Box CEO Aaron Levie shared that Fable 5.1 successfully identified subtleties and ambiguities in data that the previous version missed during testing.
Regarding benchmarks, Mythos 5.1—currently available only to Project Glasswing participants—reportedly achieves the same scores with low reasoning settings as its predecessor did at maximum reasoning levels.
Safeguards and Data Privacy
Fable 5.1 introduces "more precise safeguards" designed to be less restrictive on basic biology questions compared to Fable 5. However, Mythos 5.1 maintains the same biology restrictions as the previous iteration. Anthropic also detailed progress on data privacy through "Enterprise Frontier Safeguards," which offer complete privacy by storing data on the customer’s cloud servers rather than Anthropic’s own. This feature is slated for rollout later this fall.
In terms of specialized tasks, Anthropic is now permitting Fable 5.1 to be used for identifying software vulnerabilities. However, certain cybersecurity operations, such as penetration testing, exploit generation, and binary-based vulnerability scanning, will still be redirected to the company's Opus models.
