OpenVLM Video - Video-MME (w/o subs) - Short Videos: leaderboard
Video-MME accuracy on short videos (under 2 minutes) without subtitles, from the OpenCompass OpenVLM Video Leaderboard (VLMEvalKit): four-option multiple-choice questions pooled over all task types. Accuracy in percent, higher is better; each model is evaluated at the frame budget OpenVLM lists for it.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-78B | 83.8 | #361 |
| 2 | InternVL2.5-78B | 83 | #434 |
| 3 | InternVL3-38B | 82.1 | #413 |
| 4 | Gemini 2.0 Flash | 79.7 | #351 |
| 5 | Qwen 2 VL 72B | 78.9 | #450 |
| 6 | InternVL3-8B | 76.2 | #624 |
| 7 | Qwen2.5-Omni-7B | 75.1 | #633 |
| 8 | Qwen 2.5 VL 7B | 73.9 | #641 |
| 9 | Gemini 1.5 Flash | 71.3 | #620 |
| 10 | Qwen 2 VL 7B | 71.1 | #743 |
| 11 | MiniCPM-V-2.6 | 70.4 | #827 |
| 12 | InternVL2-8B | 65.9 | #813 |
| 13 | Claude 3.5 Sonnet | 63.6 | #362 |
| 14 | Phi-4 Multimodal Instruct | 48 | #922 |
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-video-mme-w-o-subs-short-videos · How It Works · Data refreshed daily, snapshot 2026-11-05.