Anthropic launches Claude Fable 5.1 and Mythos 5.1

Anthropic has unveiled the latest iterations of its Claude models, Claude Fable 5.1 and Claude Mythos 5.1. While Fable 5.1 is now available to the public, access to Mythos 5.1 is restricted to Anthropic’s “trusted access” programmes. According to the company, these models are, in fact, the same model with varying levels of safeguards.

Fable 5.1 is projected to be more efficient, delivering results on par with or surpassing its predecessor when set to medium or low effort. It also demonstrates heightened capabilities in agentic scientific research, achieving up to double the capability compared to its predecessor. The price for Fable 5.1 is estimated to be 25% less than Fable 5 for typical workloads, while highly agentic work could see costs reduced by up to 45%.

Additionally, Anthropic is introducing its enterprise frontier safeguards (EFS) system, which ensures customer-controlled cloud infrastructure for complete data privacy. Currently, Mythos 5.1 is accessible only to vetted professionals in cybersecurity and life sciences at select US organizations, but Anthropic is working with the US government to expand its availability.

This new version of Mythos is being utilized to power Claude Security, a tool for scanning codebases for vulnerabilities and suggesting patches for human review. The launch of these models follows recent controversies surrounding Claude’s AI capabilities in cybersecurity, leading to scrutiny from the US government. In response, OpenAI has announced the development of its Astra AI model, which will also focus on cybersecurity, though with more limited capabilities compared to Anthropic’s offerings.