Prim - Discovery: leaderboard
Metric: Discovery score (%): how well the model states the problem's mathematical primitive (the key organizing idea) from the problem alone, judged against the human-verified gold primitive by GPT-5.4-High on a graded rubric, over 182 problems; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (High) | 82.42 |
| 2 | GPT-5.4 Mini (High) | 59.34 |
| 3 | GPT-5.4 Nano (High) | 41.76 |
| 4 | GPT-OSS-20B (High) | 34.62 |
| 5 | Qwen 3.5 27B | 28.57 |
| 6 | Qwen 3.6 27B | 24.73 |
| 7 | Qwen 3.5 9B | 13.19 |
| 8 | DeepSeek-R1-Distill-Qwen-7B | 7.14 |
| 9 | Qwen 3.5 4B | 6.59 |
| 10 | DeepSeek R1 0528 Qwen3 8B | 6.59 |
| 11 | DeepSeek R1 Distill Qwen 32B | 6.04 |
| 12 | DeepSeek R1 Distill Qwen 14B | 4.4 |
Interactive version: theaggregate.ai/benchmark?slug=prim-discovery · How It Works · Data refreshed daily, snapshot 2026-10-04.