OpenVLM Video - MVBench - Object Shuffle: leaderboard
MVBench Object Shuffle accuracy from the OpenCompass OpenVLM Video Leaderboard (VLMEvalKit): one of MVBench's 20 temporal tasks, 200 multiple-choice questions on where an object ends up after being moved around. Accuracy in percent, higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-78B | 44 | #361 |
| 2 | InternVL3-38B | 42.5 | #413 |
| 3 | Phi-4 Multimodal Instruct | 41.5 | #922 |
| 4 | InternVL3-8B | 41 | #624 |
| 5 | Qwen 2 VL 7B | 39.5 | #743 |
| 6 | InternVL2.5-78B | 39.5 | #434 |
| 7 | Qwen2.5-Omni-7B | 39 | #633 |
| 8 | Qwen 2 VL 72B | 38 | #450 |
| 9 | Gemini 2.0 Flash | 37 | #351 |
| 10 | Qwen 2.5 VL 7B | 36 | #641 |
| 11 | MiniCPM-V-2.6 | 35.5 | #827 |
| 12 | Gemini 1.5 Flash | 35 | #620 |
| 13 | InternVL2-8B | 29.5 | #813 |
| 14 | Claude 3.5 Sonnet | 23 | #362 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-video-mvbench-object-shuffle · How It Works · Data refreshed daily, snapshot 2026-11-05.