Long-video understanding on M3-Bench robot
47.8MDRPyraVid
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PyraVidModel Size=32B2026.05 | 47.8 | 50 | 44.4 | 60.9 | 36.3 | 46.7 | |
| PyraVidModel Size=8B2026.05 | 43 | 35 | 39.1 | 56.1 | 26.8 | 40.9 | |
| M3-AgentModel Size=32B, Variant=RL2026.05 | 32.8 | 29.4 | 31.2 | 43.3 | 19.1 | 30.7 | |
| Gemini-2.0-FlashVariant=Flash2026.05 | 30.4 | 35 | 30.8 | 34.4 | 28.7 | 31.5 | |
| Qwen3-VL-8B-InstructModel Size=8B, Variant=Instruct2026.05 | 25.9 | 25 | 24.2 | 28.5 | 25.5 | 25 | |
| MA-LMM2026.05 | 25.6 | 23.4 | 22.7 | 39.1 | 14.4 | 24.4 | |
| Flash-VStream2026.05 | 21.6 | 19.4 | 19.3 | 24.3 | 14.1 | 19.4 | |
| MovieChat2026.05 | 13.3 | 9.8 | 12.2 | 15.7 | 7 | 11.2 |