Flat-Pack Bench - Part Mating: leaderboard
Metric: Mate: decide whether two parts are connected in the final assembly, 87 questions; multiple-choice accuracy (%, regex exact match) on furniture-assembly videos from IMaW with one or two segmented visual-prompt frames, zero-shot with greedy decoding (GPT-5 at its default); each row is the model's best of up to six settings (key-frame or trimmed video, mixed-media, collage or concat visual prompt) chosen by overall micro-average accuracy, the paper's main-table protocol; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 VL 4B (Thinking) | 59.77 |
| 2 | Qwen 3 VL 4B Instruct | 56.32 |
| 3 | Qwen 2.5 VL 32B Instruct | 54.02 |
| 4 | InternVL3-38B | 52.87 |
| 5 | Gemini 3.1 Pro (Preview) | 49.43 |
| 6 | GPT-5 | 49.43 |
| 7 | Qwen 3 VL 30B A3B Instruct | 49.43 |
| 8 | InternVL3-14B | 48.28 |
| 9 | Qwen 3 VL 32B (Thinking) | 47.13 |
| 10 | Qwen 3 VL 8B (Thinking) | 44.83 |
| 11 | Qwen 3 VL 235B A22B Instruct | 43.68 |
| 12 | Qwen 3 VL 32B Instruct | 42.53 |
| 13 | Qwen 2.5 VL 7B Instruct | 41.38 |
| 14 | Gemini 2.5 Flash | 40.23 |
| 15 | Gemini 2.5 Pro | 39.08 |
Interactive version: theaggregate.ai/benchmark?slug=flat-pack-bench-part-mating · How It Works · Data refreshed daily, snapshot 2026-10-07.