AIBIOTECHNOLOGY
Published:June 17, 2026
Affected Company:OpenAI
OpenAI Launches LifeSciBench to Test AI on Real Life-Science Research
OpenAI's new LifeSciBench benchmark, built by 173 scientists, tests whether AI can handle real life-science research work.
OpenAI has introduced LifeSciBench, a benchmark built to measure how well AI handles real-world life-science research rather than tidy fact-recall questions. Every task is grounded in the judgment of practicing scientists with Ph.D.-level training and drug-discovery experience.
It spans 750 expert-authored tasks across seven workflows and seven biological domains, drawing on 173 scientist contributors, over 19,000 rubric criteria, and 453 reviewers. The goal: test messy realities like conflicting results and decisions under uncertainty.