Anthropic Unveils Claude Fable 5.1: Cheaper and Without the Excess Restrictions

Anthropic has introduced its new Fable 5.1 and Mythos 5.1 models in an attempt to address three customer grievances: price, data storage, and overly cautious restrictions.

The company's headline pitch is cost. According to Anthropic, Fable 5.1 outperforms its predecessor while running roughly 25% cheaper in ordinary scenarios and up to 45% cheaper on complex agentic tasks. The savings come from a reduced price on data that has already been processed and stored in the cache.

First Impressions

Early takes arrived alongside the announcement. Every's head Dan Shipper called Fable 5.1 the strongest model for programming they had used, adding that it is now also fast, token-efficient, and “finally talks like a normal person.” Box CEO Aaron Levie noted that an agent running on Fable 5.1 picked up on subtleties and ambiguities in the data that the previous version had missed on the same test. On the benchmarks, meanwhile, Mythos 5.1 at a low reasoning level scores as much as its predecessor did at the maximum.

Restrictions and Privacy

Fable 5.1 has been given “more precise safeguards”: according to Anthropic, it blocks basic biology questions less often than Fable 5 did. Mythos 5.1, on the other hand, keeps the same biological restrictions.

Separately, the company described its progress on data storage. The Enterprise Frontier Safeguards mechanism promises “full privacy” by storing data on the customer's own cloud servers rather than Anthropic's. The rollout will begin closer to the end of the fall.

Also noteworthy is the softening of its cybersecurity policy: Anthropic now allows Fable 5.1 to be used for finding vulnerabilities in software. That said, some tasks will still be routed to the Opus models — including penetration testing, exploit development, and scanning for vulnerabilities in binary files.

Fable 5.1 is already available across all platforms, whereas Mythos 5.1 is so far open only to participants in the Project Glasswing program.

Source: The Verge