Anthropic Unveils BioMysteryBench as Claude Mythos Shows Promise in Biological Research

Anthropic has introduced BioMysteryBench to assess AI’s ability to tackle difficult problems in bioinformatics and life-sciences research. Such evaluations test more than factual knowledge: they examine whether models can propose research avenues that humans have yet to explore, making them an important gauge of whether AI has research-grade capabilities.

Anthropic’s latest results showed that its flagship Claude Mythos model solved nearly 30% of the biology problems that had stumped human experts. As of July 20, 2026, the findings were viewed as an important advance in AI’s move into biological research, though Anthropic had not disclosed any commercial value or data on translation into real-world scientific research.