Atlas

Benchmarks

← All benchmarks

BioMysteryBench - Human Solvable

Science · 2026-07-24

BioMysteryBench Human Solvable contains problems that independent human experts were able to solve. This is the revised subset reported from the Claude Opus 5 system card onward, after Anthropic removed 3 problems on external feedback; earlier original-set results use biomysterybench-human-solvable-original-set.

Top models (higher is better)

ModelScore
Claude Opus 590.1
Claude Fable 589.0
Gemini 3.8 Flash88.8
Opus 4.888.5
Sonnet 587.5
Gemini 3.7 Flash87.1
Loading Atlas data…