PuzzleMate - Piece Selection with Natural Language Reference (GoPro): leaderboard
A task of PuzzleMate (Dasgupta et al., 2026), an egocentric jigsaw-puzzle benchmark for multimodal assistants built on three 48-piece puzzles: the model names the piece that fits a missing slot among five unmarked candidates in a short natural-language description, on photos from a high-resolution GoPro camera. Scored as accuracy in percent as judged by human annotators; higher is better.
Metric: Accuracy (%; correct as judged by human annotators for clarity and spatial accuracy; task: naming the piece that fits a missing slot among five unmarked candidates in a natural-language reference of at most 35 tokens; three 48-piece jigsaw puzzles photographed from an egocentric viewpoint, GoPro Hero10 camera (high-resolution images)). Source: arxiv.org. Saturation forecast: Around August 2028. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 56 | #172 |
| 2 | Gemini 2.5 Pro | 54 | #262 |
| 3 | Gemma 3 27B | 46 | #1638 |
| 4 | Gemma 3 12B | 44 | #2013 |
| 5 | Gemma 3 4B | 42 | #5104 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=puzzlemate-piece-selection-with-natural-language-reference-gopro · How It Works · Data refreshed daily, snapshot 2026-11-12.