PhysCount - Correspondence 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 intersections and multiplicities of the appearance-entity assignment, such as objects seen both directly and in a mirror (lower is better).
Metric: Correspondence normalized absolute error (%). Source: arxiv.org. Saturation forecast: Around 2030. 40 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.7 Flash | 33.88 | #24 |
| 2 | Gemma 4 31B (IT) | 41.86 | #178 |
| 3 | Muse Glimmer 30B | 43.04 | #191 |
| 4 | Gemma 4 26B A4B (IT) | 44.02 | #293 |
| 5 | Qwen 3.8 27B | 44.68 | #123 |
| 6 | Gemma 4 12B (IT) | 45.7 | #332 |
| 7 | Qwen 3.5 397B A17B | 46.73 | #154 |
| 8 | GPT-6 Astra | 47.83 | #6 |
| 9 | Qwen 3.5 27B | 48.56 | #192 |
| 10 | Qwen 3.5 122B A10B | 49 | #184 |
| 11 | GPT-5.6 Terra | 50.99 | #40 |
| 12 | Qwen 3.5 35B A3B | 51.51 | #268 |
| 13 | GPT-5.6 Sol | 52.11 | #22 |
| 14 | Qwen 3.5 9B | 53.38 | #391 |
| 15 | GPT-5.6 Luna | 55.98 | #79 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=physcount-correspondence-nae · How It Works · Data refreshed daily, snapshot 2026-11-05.