LifeSciBench: Realistic, Expert-Level Life Science Tasks
OpenAI's 750 expert-authored tasks spanning 7 scientific workflows × 7 life-science domains, each with an expert rubric — built to capture the ambiguity and judgment calls that knowledge-QA benchmarks miss (complex artifacts, situational ambiguity, open-ended answers). Far from saturated: no model passes 22.8% of tasks, and 34.8% have a best-model pass rate under 20%. Score shown is the problem-weighted normalized rubric score; top pass rate is 36.1% (GPT-Rosalind).
Editor's note
OpenAI's 750 expert-authored tasks spanning 7 scientific workflows × 7 life-science domains, each with an expert rubric — built to capture the ambiguity and judgment calls that knowledge-QA benchmarks miss (complex artifacts, situational ambiguity, open-ended answers). Far from saturated: no model passes 22.8% of tasks, and 34.8% have a best-model pass rate under 20%. Score shown is the problem-weighted normalized rubric score; top pass rate is 36.1% (GPT-Rosalind).
Results
| Model | Score | Variant | Provenance |
|---|---|---|---|
| GPT-Rosalind | 57.6% norm. rubric | default | paper |
| GPT-5.5 | 51.9% norm. rubric | default | paper |
| Gemini 3.1 Pro | 51.5% norm. rubric | default | paper |
| GPT-5.4 | 47.9% norm. rubric | default | paper |
| Grok 4.3 | 39.9% norm. rubric | default | paper |
Sources
- paper LifeSciBench preprint (OpenAI · Tacit Labs)