Benchmark

LifeSciBench

Evaluate models on practical scientific reasoning and life-science research work.

Life sciences Biology Drug discovery Scientific reasoning

Source updated Jun 17, 2026Checked Oct 20, 2026OpenAI and Tacit Labs

Explore the original evidence

Explore the original evaluation, its methods, and reported findings.

Coverage

750 expert-authored tasks spanning seven workflows and seven biological domains.

Methodology

Expert-written rubrics measure average reward and task pass rate, using a 70% task-level success threshold.