PetriBench - MinimumTokenSteps (Hard): leaderboard
Exact-match accuracy on PetriBench's minimum-token-steps questions at the Hard structural-complexity level, one of six task families over Petri-net state spaces. Reasoning effort is part of the row identity. Higher is better. Source: arxiv.org/abs/2609.19883.
Metric: Exact-Match Accuracy (%). Source: arxiv.org. Saturation forecast: Around February 2027. 36 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.6 (High) | 76.8 | #29 (Grok 4.6) |
| 2 | Grok 4.6 (xHigh) | 74 | #29 (Grok 4.6) |
| 3 | Grok 4.5 (High) | 71.8 | #54 (Grok 4.5) |
| 4 | Grok 4.6 (Medium) | 71 | #29 (Grok 4.6) |
| 5 | Grok 4.5 (Medium) | 66.8 | #54 (Grok 4.5) |
| 6 | Claude Opus 5 (xHigh) | 54.8 | #16 (Claude Opus 5) |
| 7 | GPT-5.6 Sol (xHigh) | 53.2 | #22 (GPT-5.6 Sol) |
| 8 | Gemini 3.1 Pro (Preview) (High) | 50 | #64 (Gemini 3.1 Pro (Preview)) |
| 9 | Claude Opus 5 (High) | 46.5 | #16 (Claude Opus 5) |
| 10 | Gemini 3.8 Flash (Medium) | 36.8 | #30 (Gemini 3.8 Flash) |
| 11 | Gemini 3.1 Pro (Preview) (Medium) | 35.5 | #64 (Gemini 3.1 Pro (Preview)) |
| 12 | GPT-5.6 Sol (High) | 33.8 | #22 (GPT-5.6 Sol) |
| 13 | Claude Opus 5 (Medium) | 26.8 | #16 (Claude Opus 5) |
| 14 | GPT-5.6 Terra (xHigh) | 18.8 | #40 (GPT-5.6 Terra) |
| 15 | DeepSeek V4 Flash (High) | 17.2 | #155 (DeepSeek V4 Flash) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=petribench-minimumtokensteps-hard · How It Works · Data refreshed daily, snapshot 2026-11-05.