Introducing Claude Fable 5

Anthropic has released Claude Fable 5 to the public. It is the most capable model the company has ever shipped, and is positioned as the first model to combine Mythos-class capabilities with safety measures suitable for general use.

The previous-generation Mythos preview was capable enough to detect thousands of cybersecurity vulnerabilities, and Anthropic declined to release it publicly out of concern that the same capability could be repurposed for attacks. It was then offered on a limited basis to security experts protecting the world's critical software infrastructure and used to fix vulnerabilities.

Fable 5 ships with the most carefully designed safety system in the Claude series to date. Requests touching high-risk domains such as cybersecurity and biology are automatically detected and reviewed.

Requests judged to be high-risk are redirected to Opus 4.8. This lets users harness Fable's exceptional capabilities while preventing uses that carry cyberattack or biological risk. The safeguards are currently set broadly, but are slated for continual refinement so that safe requests can be better allowed.

Fable 5 is designed for ambitious work and can stay engaged with a problem far longer than previous models. It has high autonomy and can operate for days at a time without human intervention.

Beyond coding, it can handle complex tasks in finance, research, economics, and law, areas that previously always required human supervision. Anthropic urged users to 'take on problems that, in hindsight, make you wonder why they took so many years to solve,' signaling its hope that users will unlock the model's potential.