GeneBench-Pro: Multistage Statistical Reasoning in Genomics & Biology
OpenAI's June 2026 benchmark for scientific 'research taste': 129 synthetic genomics / quantitative-biology / translational-medicine problems, each pairing a noisy dataset with a target estimand, graded deterministically from a known causal structure. Extremely hard and unsaturated.
Editor's note
OpenAI's June 2026 benchmark for scientific 'research taste': 129 synthetic genomics / quantitative-biology / translational-medicine problems, each pairing a noisy dataset with a target estimand, graded deterministically from a known causal structure. Extremely hard and unsaturated.
Published results
Results are separated by evaluation version and source configuration. Scores in different sets are not a single comparable ranking.
Pro v13 · OpenAI report
Pro v13 results are separate from the June 2026 original report. Values follow the current primary report, not a secondary catalog. Task counts, answer-format validity and uncertainty are not inferred.
| Model | Score | Variant | Source |
|---|---|---|---|
| GPT-6 Astra | 37.1% reported score | Best reported effort | OpenAI · GPT-6 Astra report Science and Health table; footnote 12 |
| GPT-5.6 Sol | 32.3% reported score | Best reported effort | OpenAI · GPT-6 Astra report Science and Health table; footnote 12 |
Original paper results
| Model | Score | Variant | Source |
|---|---|---|---|
| GPT-5.6 Sol Pro | 31.5% accuracy | max reasoning | OpenAI — GeneBench-Pro report 2026-06-30 |
| GPT-5.6 Sol | 28.7% accuracy | default | OpenAI — GeneBench-Pro report 2026-06-30 |
| Claude Opus 4.8 | 16.0% accuracy | default | OpenAI — GeneBench-Pro report 2026-06-30 |
| Gemini 3.5 Flash | 8.1% accuracy | default | OpenAI — GeneBench-Pro report 2026-06-30 |