Anthropic has updated its line of neural networks, introducing the Claude Fable 5.2 and Claude Mythos 5.2 models. They are focused on solving the most complex technical tasks and during testing were able to independently design electronic devices, write an operating system from scratch, and solve open problems in black hole physics.
Anthropic has updated its line of neural networks, introducing the Claude Fable 5.2 and Claude Mythos 5.2 models. They are focused on solving the most complex technical tasks and during testing were able to independently design electronic devices, write an operating system from scratch, and solve open problems in black hole physics.
As the official developer blog notes , it's technically the same model. The difference is that Fable has strict safeguards in the areas of cybersecurity and biology, so it's available to all users on cloud platforms. Mythos, on the other hand, is only accessible to research organizations that undergo special verification.
The new models have a special emphasis on programming. The neural network already generates over 90% of changes in the base code of the company Anthropic itself. During testing, a team of agents based on Fable 5.2 successfully translated a critical service of the company from Python to Rust, accelerating its launch by 26 times. In addition, at the request of one of the engineers, artificial intelligence created a full-fledged operating system in eight days: the algorithm coordinated over 1000 subagents, writing the kernel, drivers, file storage system and graphical interface.
In addition to writing software, the models demonstrated the ability to design hardware. The artificial intelligence developed three working electronic devices based on a single text query: a home voice hub, a USB speakerphone, and a virtual pet. The neural network independently prepared specifications, selected components, designed four-layer boards and 3D case models, and each board was ready for production in less than a day.
“To ensure that security keeps pace with the growing capabilities of the models, we have conducted rigorous testing, which included developing new methods to assess specific threats, such as sandbox exits. We believe that the risks of non-compliance remain low, and the Fable 5.2 safeguards create fewer unnecessary lockouts,” Anthropic emphasizes.
The new versions generate responses 30% faster than their predecessors. The base cost is $8 per million tokens at input and $40 at output. The developers have also added an optional “fast mode” that speeds up generation by 2.5 times, but it will cost twice as much.
