Introducing LifeSciBench
OpenAI, Wednesday, June 17th, 2026
OpenAI releases LifeSciBench, a 750-task benchmark grading AI on real-world life-science research.
OpenAI introduces LifeSciBench, a benchmark for measuring how well AI supports real-world life science research. Developed with 173 PhD-level scientists and validated by 453 reviewers, it spans 750 expert-authored tasks across seven biological workflows, with 1,062 attached artifacts.
Rather than trivia, tasks mirror a principal investigator's judgment-heavy work, with 79% requiring multiple reasoning steps. The strongest model evaluated passed only 36.1%, with GPT-Rosalind improving over GPT-5.5's 25.7%.