LifeSciBench: Realistic, Expert-Level Life Science Tasks
Expert-authored life-science tasks with rubric-based scoring, designed to capture ambiguity and scientific judgment. The original 750-task preprint scores and OpenAI's later Gold v1 results are separate; a rubric score is not a whole-task pass rate.
Editor's note
Expert-authored life-science tasks with rubric-based scoring, designed to capture ambiguity and scientific judgment. The original 750-task preprint scores and OpenAI's later Gold v1 results are separate; a rubric score is not a whole-task pass rate.
Published results
Results are separated by evaluation version and source configuration. Scores in different sets are not a single comparable ranking.
Gold v1 · OpenAI report
OpenAI's later Gold v1 evaluation is separate from the original preprint. Reported values use the best score at any effort. Task counts, answer-format validity and uncertainty are not specified in this table; none are inferred.
| Model | Score | Variant | Source |
|---|---|---|---|
| GPT-6 Astra | 60.3% reported score | Best reported effort | OpenAI · GPT-6 Astra report Science and Health table; footnote 12 |
| GPT-5.6 Sol | 59.9% reported score | Best reported effort | OpenAI · GPT-6 Astra report Science and Health table; footnote 12 |
Original paper results
| Model | Score | Variant | Source |
|---|---|---|---|
| GPT-Rosalind | 57.6% norm. rubric | default | LifeSciBench preprint (OpenAI · Tacit Labs) |
| GPT-5.5 | 51.9% norm. rubric | default | LifeSciBench preprint (OpenAI · Tacit Labs) |
| Gemini 3.1 Pro | 51.5% norm. rubric | default | LifeSciBench preprint (OpenAI · Tacit Labs) |
| GPT-5.4 | 47.9% norm. rubric | default | LifeSciBench preprint (OpenAI · Tacit Labs) |
| Grok 4.3 | 39.9% norm. rubric | default | LifeSciBench preprint (OpenAI · Tacit Labs) |