Flat-Pack Bench - Temporal Ordering: leaderboard
Metric: TOrd: order the connection events of the highlighted parts, 155 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: Around 2033. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | InternVL3-78B | 43.87 |
| 2 | InternVL3-38B | 42.58 |
| 3 | InternVL3-14B | 42.58 |
| 4 | Qwen 2.5 VL 72B Instruct | 41.29 |
| 5 | Gemini 2.5 Pro | 40.65 |
| 6 | GPT-5 | 40.65 |
| 7 | Qwen 3 VL 32B Instruct | 38.71 |
| 8 | Qwen 3 VL 32B (Thinking) | 38.71 |
| 9 | Qwen 3 VL 235B A22B Instruct | 37.42 |
| 10 | Qwen 3 VL 8B Instruct | 36.13 |
| 11 | Gemini 3.1 Pro (Preview) | 34.84 |
| 12 | Qwen 2.5 VL 32B Instruct | 34.84 |
| 13 | Qwen 3 VL 4B Instruct | 34.19 |
| 14 | Qwen 3 VL 8B (Thinking) | 34.19 |
| 15 | Gemini 2.5 Flash | 31.61 |
Interactive version: theaggregate.ai/benchmark?slug=flat-pack-bench-temporal-ordering · How It Works · Data refreshed daily, snapshot 2026-10-07.