New models improved results in programming, logical reasoning, cybersecurity, and robotics.

Fable 5.2 formalized the proof of Fermat's Last Theorem in 30 hours using a team of 1,000 agents.

Mythos 5.2 detects more software vulnerabilities but is available only to verified organizations.

Anthropic presented updated models Claude Fable 5.2 and Claude Mythos 5.2, designed to perform complex technical tasks. According to the company, they demonstrate improvements in programming, logical reasoning, and computer vision.

During preliminary testing, the models helped engineers carry out a complex software migration, create three working electronic devices, solve two open problems in black hole physics, and advance research in robotics.

One of the most notable results was the formalization of the proof of Fermat's Last Theorem. A team of 1,000 agents based on Fable 5.2 completed the task in 30 hours with minimal human intervention.

Both models are built on the same base architecture but have different access conditions. Fable 5.2 received strengthened safeguards in the areas of cybersecurity and biology, which makes it possible to make it available to a wide range of users. Mythos 5.2 is intended for verified organizations working on research in the fields of cybersecurity and life sciences.

Programming, physics, and robotics

According to Anthropic, more than 90% of changes to the company's main codebase are already created by Claude under human supervision.

In one test, Fable 5.2 coordinated the work of more than 1,000 subagents who developed an operating system from scratch over eight days. The result was a system that boots on a laptop and has its own kernel, drivers, file storage, window interface, and installer.

In another experiment, a team of agents based on the model ported a critical production service from Python to Rust. According to the company's estimate, after the migration it became approximately 26 times faster.

In scientific research, Fable 5.2 solved two open problems related to black holes and holographic duality. The model also demonstrated the ability to work with mathematical proofs and complex physics calculations.

In the field of robotics, Fable 5.2 showed high results in the RoboDojo benchmark, which tests the ability of models to control a simulated two-armed robot based on camera images.

In addition, the model independently developed three electronic devices: a home hub with a voice assistant, a USB speakerphone, and a pocket virtual pet. It prepared specifications, selected components, designed printed circuit boards and enclosures, and also wrote the software.

Mythos 5.2 strengthened cybersecurity capabilities

Anthropic called Mythos 5.2 its most powerful model for engineering tasks in the field of cybersecurity. According to the company's internal testing, it detected 10.2% more vulnerabilities than Claude Opus 5.5 and 14.4% more than the previous version, Mythos 5.1.

One company specializing in cybersecurity reported that other tested advanced models missed every sixth confirmed vulnerability found by Mythos 5.2.

At the same time, access to these capabilities is restricted. Organizations must undergo verification within the Cyber Verification Program. Corporate clients can also use Mythos 5.2 through the Claude Security tool to search for vulnerabilities in software code.

Cost and availability

Anthropic reported that the new models generate responses up to 30% faster than previous versions. The company also reduced their cost.

Both models have an accelerated mode that can increase token generation speed by up to 2.5 times. Access to Mythos's fast mode is also restricted to participants in verification programs.

The company recommends using Fable and Mythos for the most complex and large-scale technical tasks. For everyday work with code and texts, Anthropic advises starting with Claude Opus 5.5 and, if necessary, moving to more powerful or lighter models.

The company also stated that during testing of the new models they did not find a critical increase in the risks of dangerous behavior. According to the results of internal evaluations, alignment indicators were the same as or better than those of previous models. At the same time, Anthropic acknowledges that no single test can fully predict a model's behavior after deployment.

Recall that in October 2026, Anthropic filed for an IPO on the Nasdaq exchange.