OpenAI Releases LifeSciBench: Measuring AI Systems’ Capabilities in Real-World Scientific Research Scenarios

TechFlow reports that on June 18, OpenAI released LifeSciBench, a new evaluation benchmark designed to assess AI systems’ capabilities in real-world scientific research scenarios. LifeSciBench comprises 750 expert-crafted tasks covering seven categories of scientific workflows and seven domains of biology. The tasks were contributed by 173 researchers holding doctoral degrees and possessing experience in biotechnology or pharmaceutical industries. Emphasizing the assessment of complex scientific research capabilities—including evidence integration, experimental design, data analysis, scientific reasoning, and scientific communication—rather than isolated factual questions, over 79% of the tasks require multi-step reasoning, with an average of approximately four reasoning steps per task. The benchmark also includes 1,062 authentic research-related data attachments (e.g., research papers, figures, sequence data, and structural files).