Anthropic is pitching its latest Claude releases as an answer to customers who want capable AI without punishing costs, intrusive data handling or overly eager guardrails. The company’s balancing act is clear: broaden Fable’s usefulness while keeping its more sensitive Mythos model tightly controlled.

On Tuesday, Anthropic released Fable 5.1 and Mythos 5.1, presenting the paired models as upgrades in performance and efficiency. Fable is available through Anthropic’s API and cloud platforms, while Mythos remains restricted to registered partners working in cybersecurity or life-sciences research.¹

The commercial pitch centers on cheaper agentic work. Anthropic says Fable 5.1 is typically about 25% less expensive than its predecessor and can be up to 45% cheaper on complex tasks because cached data costs less to process.² A repost from developer advocate Addy Osmani highlighted the same shift: cache reads are “75% cheaper,” while the model can stay with long tasks longer and communicate more naturally when it gets stuck.Addy Osmani (addyosmani)³

Anthropic is also responding to complaints that safety filters can get in the way of ordinary work. Fable 5.1 has more precise safeguards intended to reduce false-positive blocks, including on basic biology questions, and is now permitted to help identify software vulnerabilities. Yet the company will still redirect higher-risk tasks—such as exploit generation and penetration testing—to Opus models.²

Privacy is the other major promise. Enterprise Frontier Safeguards, due to begin rolling out in the fall, would let customers keep data on their own cloud infrastructure while retaining misuse monitoring under customer control. Anthropic said it has “never trained on enterprise data without explicit permission, and never will.”¹

The company’s own safety assessment nonetheless flags a compromise. Mythos 5.1 is described as a slight regression from Opus 5 on overall misaligned behavior, even as it improves on earlier Mythos and Sonnet versions—an admission that greater capability does not erase governance risks.¹