Long-video understanding on LongVideoBench (LVB)
76.8AccuracyGemini 1.5 Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini 1.5 ProSource=[60]2026.06 | 76.8 | |
| Gemini 1.5 ProSource=[59]2026.06 | 75.9 | |
| Gemini 1.5 Flash2026.06 | 73.1 | |
| GPT-4o2026.06 | 72.6 | |
| GPT-4o mini2026.06 | 69.7 | |
| AdaCodecToken budget=comparable, Backbone=Qwen2-VL-7B2026.06 | 67.8 | |
| Molmo2-8B2026.06 | 67.5 | |
| Eagle-X5-7B2026.06 | 66.4 | |
| Key-VL-1.5-8B2026.06 | 66 | |
| GLM-4V-9B2026.06 | 65.7 | |
| Claude 3.5 Sonnet2026.06 | 65.1 | |
| VideoChat-Flash-7B2026.06 | 64.7 | |
| MiniCPM-V-2.6-8B2026.06 | 63.9 | |
| Molmo2-O-7B2026.06 | 63.7 | |
| AdaCodecToken budget=1/7, Backbone=Qwen2-VL-7B2026.06 | 63.2 | |
| Qwen2-VL-7B2026.06 | 62.4 | |
| InternVL2-8B2026.06 | 62.1 | |
| ReMoRa-7B2026.06 | 60.8 | |
| LLaVA-Video-7B2026.06 | 58.2 | |
| PLM-7B2026.06 | 56.9 | |
| CoPE-VideoLM-7B2026.06 | 56.9 |