Do You See Me - 2D Perception: leaderboard
Each model answers 2,116 generated 2D puzzles across seven perception subtasks (figure-ground, spatial grids, colour, shape and letter disambiguation, form constancy, visual closure); the score is the mean of the subtask accuracies. Unlike MMVP or CV-Bench, difficulty is set by generator parameters rather than by collected photos.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 50.09 | #648 |
| 2 | Qwen 2.5 VL 7B Instruct | 34.7 | #1893 |
| 3 | Gemma 3 12B (IT) | 28.63 | #1891 |
| 4 | InternVL2.5-8B | 26.58 | #2401 |
| 5 | Phi-4 Multimodal Instruct | 17.63 | #4209 |
| 6 | Llama 3.2 11B Instruct | 16.84 | #5226 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=do-you-see-me-2d-perception · How It Works · Data refreshed daily, snapshot 2026-11-17.