PetriBench - Hard: leaderboard
Exact-match accuracy on PetriBench's Hard tier, where accuracy drops most and task-specific capability profiles separate. 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 October 2027. 36 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (xHigh) | 77.8 | #22 (GPT-5.6 Sol) |
| 2 | Claude Opus 5 (High) | 71.8 | #16 (Claude Opus 5) |
| 3 | Claude Opus 5 (xHigh) | 71.7 | #16 (Claude Opus 5) |
| 4 | GPT-5.6 Sol (High) | 59.8 | #22 (GPT-5.6 Sol) |
| 5 | Claude Opus 5 (Medium) | 58.1 | #16 (Claude Opus 5) |
| 6 | Grok 4.6 (xHigh) | 56.5 | #29 (Grok 4.6) |
| 7 | Grok 4.6 (High) | 55 | #29 (Grok 4.6) |
| 8 | Grok 4.6 (Medium) | 52.6 | #29 (Grok 4.6) |
| 9 | GPT-5.6 Terra (xHigh) | 50.7 | #40 (GPT-5.6 Terra) |
| 10 | Grok 4.5 (High) | 50.1 | #54 (Grok 4.5) |
| 11 | Grok 4.5 (Medium) | 48.9 | #54 (Grok 4.5) |
| 12 | Gemini 3.1 Pro (Preview) (High) | 46.9 | #64 (Gemini 3.1 Pro (Preview)) |
| 13 | Gemini 3.8 Flash (High) | 41.3 | #30 (Gemini 3.8 Flash) |
| 14 | Gemini 3.8 Flash (Medium) | 38.5 | #30 (Gemini 3.8 Flash) |
| 15 | Gemini 3.1 Pro (Preview) (Medium) | 34.8 | #64 (Gemini 3.1 Pro (Preview)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=petribench-hard · How It Works · Data refreshed daily, snapshot 2026-11-05.