Anthropic Releases BioMysteryBench Benchmark

Benchmark Stats

Release Date
2026-04-29
Total Problems
99
Human-Difficult Problems
23
Mythos Preview Solve Rate (Difficult)
30%

Anthropic released BioMysteryBench, a new benchmark with 99 bioinformatics problems from real datasets. Latest Claude models including Mythos Preview solved around 30% of the 23 problems that stumped human experts.