Long-form video understanding and instruction following on LVBench (test)
78.4AccuracyGemini-2.5-Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-2.5-ProAccess=Proprietary2026.02 | 78.4 | |
| Gemini-3-ProAccess=Proprietary2026.02 | 78 | |
| GPT-5Access=Proprietary2026.02 | 68.8 | |
| Claude-Sonnet-4.5Access=Proprietary2026.02 | 50.5 | |
| CoPE-VideoLMModel scale=7B2026.02 | 46.4 | |
| LLaVA-Video-7BModel scale=7B2026.02 | 44.2 | |
| mPLUG-Owl3-7BModel scale=7B2026.02 | 43.5 | |
| TimeMarker2026.02 | 41.3 | |
| Kangaroo-8BModel scale=8B2026.02 | 39.4 | |
| LLaVA-OV-7BModel scale=7B2026.02 | 38.1 | |
| PLLaVA-34BModel scale=34B2026.02 | 26.1 |