SGI-Bench Idea Generation: leaderboard
SGI-Bench subscore for proposing plausible, useful, and novel scientific research ideas.
Metric: Idea Generation Score. Source: github.com. Saturation forecast: Around 2029. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 55.4 | #98 |
| 2 | GPT-5.2 Pro | 55.03 | #61 |
| 3 | GPT-5.1 | 47.12 | #135 |
| 4 | O3 | 46.07 | #124 |
| 5 | Claude Sonnet 4.5 | 43.2 | #142 |
| 6 | O4 Mini | 40.78 | #180 |
| 7 | Claude Opus 4.1 | 40.29 | #140 |
| 8 | Gemini 2.5 Pro | 39.95 | #152 |
| 9 | Qwen 3 Max | 39.83 | #208 |
| 10 | Gemini 3 Pro | 39.68 | #80 |
| 11 | Qwen 3 VL 235B A22B | 39.28 | #238 |
| 12 | Gemini 2.5 Flash | 39.13 | #245 |
| 13 | Intern-S1 | 38.09 | #285 |
| 14 | Grok 4 | 37.12 | #181 |
| 15 | GPT-4.1 | 36.49 | #248 |
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-idea-generation · How It Works · Data refreshed daily, snapshot 2026-10-20.