Complex Reasoning on vcrbench
53.1AccuracyVideoLatent-7B (Ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| VideoLatent-7B (Ours)Frames=32, Architecture Category=Latent MLLM2026.06 | 53.1 | |
| Open-o3-Video-7BFrames=32, Architecture Category=Standard MLLM2026.06 | 52.5 | |
| VideoLatent-7B (Ours)Frames=64, Architecture Category=Latent MLLM2026.06 | 52.5 | |
| Qwen2.5-VL-7BFrames=64, Architecture Category=Standard MLLM2026.06 | 51.4 | |
| Video-R1-7BFrames=64, Architecture Category=Standard MLLM2026.06 | 50.8 | |
| Mull-Token-7BFrames=32, Architecture Category=Latent MLLM2026.06 | 50.6 | |
| Mull-Token-7BFrames=64, Architecture Category=Latent MLLM2026.06 | 50.6 | |
| Qwen2.5-VL-7BFrames=32, Architecture Category=Standard MLLM2026.06 | 50.4 | |
| VideoLatent-7B (Ours)Frames=16, Architecture Category=Latent MLLM2026.06 | 49.6 | |
| Open-o3-Video-7BFrames=64, Architecture Category=Standard MLLM2026.06 | 49.2 | |
| LVR-7BFrames=32, Architecture Category=Latent MLLM2026.06 | 49 | |
| Mull-Token-7BFrames=16, Architecture Category=Latent MLLM2026.06 | 48.6 | |
| LVR-7BFrames=64, Architecture Category=Latent MLLM2026.06 | 48 | |
| Video-R1-7BFrames=32, Architecture Category=Standard MLLM2026.06 | 46.9 | |
| Qwen2.5-VL-7BFrames=16, Architecture Category=Standard MLLM2026.06 | 46.7 | |
| Video-R1-7BFrames=16, Architecture Category=Standard MLLM2026.06 | 46.3 | |
| LVR-7BFrames=16, Architecture Category=Latent MLLM2026.06 | 45.3 | |
| Qwen3-VL-8B + SynRLModel Scale=8B, SynRL Training=true2026.03 | 35.6 | |
| Qwen3-VL-8BModel Scale=8B, SynRL Training=false2026.03 | 34.2 | |
| Qwen3-VL-4B + SynRLModel Scale=4B, SynRL Training=true2026.03 | 31.5 | |
| Qwen3-VL-4BModel Scale=4B, SynRL Training=false2026.03 | 30.1 | |
| Monet-7BFrames=32, Architecture Category=Latent MLLM2026.06 | 25.5 | |
| Monet-7BFrames=64, Architecture Category=Latent MLLM2026.06 | 25.1 | |
| Monet-7BFrames=16, Architecture Category=Latent MLLM2026.06 | 24.7 |