ERGeoBench - Common Sense: leaderboard
Common Sense diagnostics on ERGeoBench (2026): questions that map a user need to a visible point of interest in single, dual and cross views, averaged over three sub-tasks. Accuracy in percent; higher is better.
Metric: Accuracy (%) on the common sense diagnostic questions: single-view, dual-view and cross-view questions mapping user needs to points of interest, fuzzy-matched, averaged over the three sub-tasks; unified structured prompt with egocentric observations, action history and in-context examples, temperature 0.1, first valid JSON object parsed (invalid outputs count as failures); higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 9 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.0 Flash | 60.72 | #723 |
| 2 | Qwen 3 VL 8B | 55.51 | #917 |
| 3 | InternVL3-8B | 52.85 | #1782 |
| 4 | Qwen 2.5 VL 7B | 50.55 | #1862 |
| 5 | InternVL2.5-8B | 46.22 | #2370 |
| 6 | GPT-4o | 42.38 | #751 |
| 7 | Gemini 2.5 Pro | 38.24 | #276 |
| 8 | Gemini 3 Flash | 37.02 | #175 |
| 9 | Qwen-VL-Max | 33 | #1400 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=ergeobench-common-sense · How It Works · Data refreshed daily, snapshot 2026-11-21.