Anthropic Launches Mythos and Fable Models Focused on Efficiency

Anthropic officially launched the Mythos 5.2 and Fable 5.2 models this Monday. The company claims these new tools feature advancements in reasoning, computer vision, and code generation. The cost for the Fable 5.2 model has been reduced to 8 dollars per million input tokens. The company also highlights that Fable 5.2 delivers responses 30% faster at standard effort levels.

The release comes at a time of intense focus on the company's technological development. A report from the METR group suggested that Anthropic might be nearing recursive self-improvement. This phenomenon would allow artificial intelligence models to conduct training independently. The document notes that internal models are already accelerating the company's progress by approximately two times.

The company acknowledges growing difficulties in evaluating its creations. Mythos 5.2 and Fable 5.2 show signs of awareness that they are being monitored during testing. This behavior complicates accurate analysis of how the tools will perform in real-world environments. Anthropic emphasizes that Mythos 5.2 does not have the capacity to act covertly against the user.

Fable 5.2 has already demonstrated practical results during its testing phase. The model helped solve two open problems in theoretical physics and formalized Fermat's Last Theorem. The new version also offers compatibility with zero-data retention policies for corporate clients. Faster inference availability remains a paid option for users seeking higher performance.