PhysCount - Group-Acc: leaderboard
PhysCount (arXiv 2610.40530): 4,595 human-written counting questions on 546 real photos with mirrors and glass, where one physical object can appear several times by reflection or transmission. Share of the 651 annotation groups in which every appearance, entity, image-formation and correspondence answer is exactly right (higher is better).
Metric: Group accuracy (%): groups with every answer exact. Source: arxiv.org. Saturation forecast: Around 2033. 40 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.7 Flash | 13.36 | #24 |
| 2 | Muse Glimmer 30B | 7.83 | #191 |
| 3 | Qwen 3.8 27B | 7.37 | #123 |
| 4 | Gemma 4 31B (IT) | 6.61 | #178 |
| 5 | Qwen 3.5 27B | 6.3 | #192 |
| 6 | Qwen 3.5 35B A3B | 5.99 | #268 |
| 7 | Qwen 3.5 122B A10B | 5.99 | #184 |
| 8 | Qwen 3.5 397B A17B | 5.84 | #154 |
| 9 | Gemma 4 26B A4B (IT) | 5.53 | #293 |
| 10 | Qwen 3.5 9B | 5.07 | #391 |
| 11 | GPT-5.6 Terra | 4.3 | #40 |
| 12 | GPT-5.6 Sol | 3.84 | #22 |
| 13 | Qwen 3.5 4B | 3.38 | #514 |
| 14 | GPT-6 Astra | 3.38 | #6 |
| 15 | Gemma 4 12B (IT) | 3.07 | #332 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=physcount-group-acc · How It Works · Data refreshed daily, snapshot 2026-11-05.