PhysCount - Appearance 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 enumerating every visible occurrence of the target, direct, reflected or seen through glass (lower is better).
Metric: Appearance normalized absolute error (%). Source: arxiv.org. Saturation forecast: Around February 2027. 40 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 122B A10B | 17.96 | #184 |
| 2 | Qwen 3.5 35B A3B | 18.81 | #268 |
| 3 | Qwen 3.5 397B A17B | 18.99 | #154 |
| 4 | Qwen 3.8 27B | 19.89 | #123 |
| 5 | Qwen 3.5 9B | 20.04 | #391 |
| 6 | Gemini 3.7 Flash | 20.08 | #24 |
| 7 | Muse Glimmer 30B | 20.31 | #191 |
| 8 | Qwen 3.5 27B | 20.64 | #192 |
| 9 | Gemma 4 31B (IT) | 21.83 | #178 |
| 10 | Qwen 3.5 4B | 22.19 | #514 |
| 11 | Gemma 4 26B A4B (IT) | 22.7 | #293 |
| 12 | GPT-5.6 Sol | 22.82 | #22 |
| 13 | GPT-6 Astra | 24.87 | #6 |
| 14 | Gemma 4 12B (IT) | 26.05 | #332 |
| 15 | GPT-5.6 Terra | 26.06 | #40 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=physcount-appearance-nae · How It Works · Data refreshed daily, snapshot 2026-11-05.