SGI-Bench Experimental Reasoning: leaderboard
SGI-Bench subscore for reasoning about experimental setup, controls, interpretation, and scientific evidence.
Metric: Experimental Reasoning Score. Source: github.com. Saturation forecast: Around 2029. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 41.92 | #80 |
| 2 | Gemini 2.5 Pro | 41.24 | #152 |
| 3 | GPT-5.2 Pro | 39.18 | #61 |
| 4 | Claude Opus 4.1 | 38.83 | #140 |
| 5 | GPT-4.1 | 38.49 | #248 |
| 6 | GPT-5 | 38.14 | #98 |
| 7 | Claude Sonnet 4.5 | 37.8 | #142 |
| 8 | Qwen 3 Max | 37.8 | #208 |
| 9 | Gemini 2.5 Flash | 34.36 | #245 |
| 10 | GPT-5.1 | 34.02 | #135 |
| 11 | O4 Mini | 33.33 | #180 |
| 12 | O3 | 32.65 | #124 |
| 13 | GPT-4o | 32.3 | #343 |
| 14 | Qwen 3 VL 235B A22B | 31.62 | #238 |
| 15 | Grok 4 | 30.24 | #181 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sgi-bench-experimental-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.