Procedural Temporal Understanding on LEGO-Puzzles (test)
42.82Overall AccuracyTPRU-7B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TPRU-7BBackbone=7B2026.02 | 42.82 | 3,200 | |
| LLaVA-Video-Qwen2-72BBackbone=Qwen2-72B2026.02 | 41.64 | 500 | |
| Qwen2.5-Omni-7BBackbone=Qwen2.5-7B2026.02 | 36.45 | 1,200 | |
| Long-VITA-16KContext Window=16K2026.02 | 34.45 | 200 | |
| LLaVA-Video-Qwen2-7BBackbone=Qwen2-7B2026.02 | 31.91 | 100 | |
| SmolVLM2-256M-VideoParameters=256M2026.02 | 28.18 | 0 | |
| SmolVLM2-500M-VideoParameters=500M2026.02 | 27.27 | 0 |