Introducing LifeSciBench
OpenAI
LifeSciBench is a new benchmark designed to evaluate AI performance on realistic, expert-level scientific research tasks in the life sciences.Summary
LifeSciBench is a comprehensive benchmark created to assess the capabilities of AI systems in life science research. Developed with input from over 170 expert scientists, it features 750 tasks across seven domains, focusing on practical workflows like evidence handling, experimental design, and translation. Unlike traditional benchmarks that rely on simple fact-recall, LifeSciBench uses granular rubrics to evaluate whether models can perform complex, multi-step scientific reasoning and provide outputs useful for real-world industry applications.
(Source：OpenAI)