P2NPCBench - P - Family 9: leaderboard
P2NPCBench:12 final benchmark models, Python 3 single-solution pass rate, temperature 0, top-p 1, maximum 32,768 output tokens; paired polynomial and NP-complete problem statements across 12 families. Original labels retained where checkpoint or instruction suffix is not specified; hardening-panel scores are separate. Score (%).
Metric: Score (%). Source: arxiv.org. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.8 Flash | 100 | #23 |
| 2 | GPT-5.6 Sol | 95.6 | #18 |
| 3 | Gemini 3.7 Flash | 92.8 | #22 |
| 4 | Qwen 3.8 Flash | 71.3 | #55 |
| 5 | Qwen 3.8 27B | 70.6 | #118 |
| 6 | Qwen3.8 Omni Flash | 62.7 | #46 |
| 7 | DeepSeek V4.1 Flash | 60.6 | #33 |
| 8 | Mistral Medium 3.5 | 46.2 | #363 |
| 9 | Gemma 3 27B | 22.2 | #610 |
| 10 | Llama 4 Scout | 12.5 | #660 |
| 11 | Gemma 3 12B | 10.2 | #673 |
| 12 | Gemma 3 4B | 7.3 | #1102 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=p2npcbench-p-family-9 · How It Works · Data refreshed daily, snapshot 2026-10-23.