DGEval - Regulatory Recall: leaderboard
The section of the same maritime dangerous-goods exam that nothing saturates: name the IMDG Code provision a scenario falls under, from memory. Scores run from 0 to 37 percent, so it separates the frontier where the multiple-choice half no longer does.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around March 2028. 35 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 36.8 | #58 (Gemini 3.1 Pro (Preview)) |
| 2 | Gemini 3.1 Pro (Preview) (Medium) | 35 | #58 (Gemini 3.1 Pro (Preview)) |
| 3 | Gemini 3.1 Pro (Preview) (Low) | 34.1 | #58 (Gemini 3.1 Pro (Preview)) |
| 4 | GPT-5.5 (High) | 30.5 | #31 (GPT-5.5) |
| 5 | GPT-5.5 (Medium) | 27 | #31 (GPT-5.5) |
| 6 | GPT-5.5 (Low) | 25.8 | #31 (GPT-5.5) |
| 7 | GPT-5.5 (Non-reasoning) | 22 | #31 (GPT-5.5) |
| 8 | Gemini 3.1 Flash Lite (High) | 22 | #191 (Gemini 3.1 Flash Lite) |
| 9 | Claude Opus 4.7 (Max) | 20.2 | #49 (Claude Opus 4.7) |
| 10 | Gemini 3.1 Flash Lite (Medium) | 19.3 | #191 (Gemini 3.1 Flash Lite) |
| 11 | Gemini 3.1 Flash Lite (Low) | 17.9 | #191 (Gemini 3.1 Flash Lite) |
| 12 | Gemini 3.1 Flash Lite (Minimal) | 17.8 | #191 (Gemini 3.1 Flash Lite) |
| 13 | Claude Opus 4.7 (Medium) | 17.6 | #49 (Claude Opus 4.7) |
| 14 | Claude Opus 4.7 (Non-reasoning) | 17.4 | #49 (Claude Opus 4.7) |
| 15 | Claude Opus 4.7 (High) | 17.2 | #49 (Claude Opus 4.7) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=dgeval-regulatory-recall · How It Works · Data refreshed daily, snapshot 2026-10-20.