OpenAI has introduced LifeSciBench, a benchmark designed to evaluate how AI systems handle real-world life science research tasks and decisions. Built to address the limitations of narrow domain evaluations, LifeSciBench features 750 expert-authored tasks spanning seven workflows and seven biological domains, incorporating 1,062 attached artifacts such as figures, PDFs, and structural data files. The benchmark was developed by 173 expert scientists with Ph.D.-level training and biotechnology or pharmaceutical industry experience, and was further validated by 453 independent expert reviewers. Tasks are graded using granular rubrics averaging 25 criteria per task to assess scientific correctness, detail, justification, and usefulness for research decisions. OpenAI reports two metrics: pass rate based on a 73% success threshold, and score based on average rubric rewards. Evaluations show that GPT-Rosalind achieves an overall exact pass rate of 36.1%, up from 25.7% for GPT-5.5. The strongest capability progressions appear in scientific communication and translation, where GPT-Rosalind improved significantly over GPT-5.5. However, performance remains weaker on artifact-heavy, design-heavy, and operationally constrained tasks, with design and optimization and analysis remaining among the most difficult workflows for current models.
Introducing LifeSciBench
OpenAI has introduced LifeSciBench, a new benchmark designed to evaluate AI systems on real-world life science research tasks. Created with input from 173 expert scientists, the benchmark features 750 tasks across seven workflows and biological domains. Initial evaluations show that GPT-Rosalind improves upon GPT-5.5, particularly in scientific communication and translation, though significant challenges remain in artifact-heavy and design-focused tasks.
Why it matters
LifeSciBench addresses a critical gap in AI evaluation by testing complex, multi-step scientific reasoning and decision-making under uncertainty rather than isolated fact-recall. This provides the AI industry with a rigorous standard for measuring whether models can serve as effective, practical collaborators in drug discovery and applied life science research.