SciHorizon-Gene - Overall: leaderboard
Evaluates gene-centered knowledge and functional understanding in the paper's zero-shot study: 2,710 sampled questions produce 9,010 evaluation instances with cyclic option permutation. Overall is the published equal-weight task score, combining accuracy, completeness and functional tasks; it is not one uniform accuracy metric. Missing task cells are not filled, and source model labels are retained. This series reports SciHorizon-Gene - Overall using Published Mean Task Score (0-100).
Metric: Published Mean Task Score (0-100). Source: arxiv.org. Saturation forecast: Around May 2028. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 61.92 | #105 |
| 2 | Claude 3.5 Sonnet | 59.35 | #362 |
| 3 | Mistral Medium 3.1 | 59.17 | #418 |
| 4 | Gemini 2.5 Pro | 58.19 | #159 |
| 5 | Claude Sonnet 4.5 | 57.04 | #148 |
| 6 | GPT-4o | 56.12 | #357 |
| 7 | GPT-4 Turbo | 54.6 | #417 |
| 8 | Gemini 2.5 Flash | 54.15 | #255 |
| 9 | Qwen 2.5 72B | 54.02 | #512 |
| 10 | Llama 3.1 405B | 51.26 | #476 |
| 11 | Mistral Small 3.1 | 50.23 | #623 |
| 12 | Llama 3.1 70B | 49.56 | #588 |
| 13 | Gemma 3 27B | 47.75 | #606 |
| 14 | Ministral 8B | 45.98 | #1036 |
| 15 | GPT-3.5 Turbo | 45.79 | #858 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scihorizon-gene-overall · How It Works · Data refreshed daily, snapshot 2026-11-05.