PetriBench - Easy: leaderboard
Exact-match accuracy on PetriBench's Easy tier: reasoning over dynamic state spaces modelled as Petri nets (boundedness, deadlock, reachable markings, minimum token steps, liveness), evaluated against exact ground truth. 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 | Gemini 3.8 Flash (High) | 96.8 | #30 (Gemini 3.8 Flash) |
| 2 | GPT-5.6 Sol (xHigh) | 96.4 | #22 (GPT-5.6 Sol) |
| 3 | Grok 4.6 (xHigh) | 95.3 | #29 (Grok 4.6) |
| 4 | Grok 4.6 (High) | 95.1 | #29 (Grok 4.6) |
| 5 | Grok 4.6 (Medium) | 93.9 | #29 (Grok 4.6) |
| 6 | GPT-5.6 Sol (High) | 92.9 | #22 (GPT-5.6 Sol) |
| 7 | Claude Opus 5 (xHigh) | 92.7 | #16 (Claude Opus 5) |
| 8 | Gemini 3.1 Pro (Preview) (High) | 92.3 | #64 (Gemini 3.1 Pro (Preview)) |
| 9 | Claude Opus 5 (High) | 90.4 | #16 (Claude Opus 5) |
| 10 | Grok 4.5 (High) | 88.2 | #54 (Grok 4.5) |
| 11 | Gemini 3.8 Flash (Medium) | 86.6 | #30 (Gemini 3.8 Flash) |
| 12 | GPT-5.6 Terra (xHigh) | 85 | #40 (GPT-5.6 Terra) |
| 13 | Claude Opus 5 (Medium) | 84.3 | #16 (Claude Opus 5) |
| 14 | Grok 4.5 (Medium) | 83.8 | #54 (Grok 4.5) |
| 15 | GPT-5.6 Sol (Medium) | 82.7 | #22 (GPT-5.6 Sol) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=petribench-easy · How It Works · Data refreshed daily, snapshot 2026-11-05.