PhysCount - Entity NAE: 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. Normalized absolute error on counting physical objects, merging the appearances of one entity (lower is better).
Metric: Entity normalized absolute error (%). Source: arxiv.org. Saturation forecast: Around 2028. 40 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.7 Flash | 27.36 | #24 |
| 2 | Muse Glimmer 30B | 32.1 | #191 |
| 3 | Qwen 3.8 27B | 32.42 | #123 |
| 4 | Qwen 3.5 397B A17B | 33.28 | #154 |
| 5 | Qwen 3.5 35B A3B | 33.44 | #268 |
| 6 | Qwen 3.5 122B A10B | 34.4 | #184 |
| 7 | Qwen 3.5 27B | 34.58 | #192 |
| 8 | Qwen 3.5 9B | 36.09 | #391 |
| 9 | GPT-5.6 Terra | 36.39 | #40 |
| 10 | GPT-5.6 Sol | 36.58 | #22 |
| 11 | Gemma 4 31B (IT) | 36.88 | #178 |
| 12 | Molmo2-8B | 37.5 | |
| 13 | GPT-6 Astra | 37.72 | #6 |
| 14 | Gemma 4 26B A4B (IT) | 39.22 | #293 |
| 15 | Qwen 3.5 4B | 39.57 | #514 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=physcount-entity-nae · How It Works · Data refreshed daily, snapshot 2026-11-05.