OpenVLM Video - TempCompass - Attribute Change: leaderboard
TempCompass Attribute Change accuracy from the OpenCompass OpenVLM Video Leaderboard (VLMEvalKit): questions on how attributes such as size or colour change in videos where a single frame cannot answer, pooled over the multi-choice, yes-or-no, caption-matching and captioning formats. Percent, higher is better; runs with a failed format (exactly 0) are left out.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2 VL 72B | 86.29 | #450 |
| 2 | InternVL3-38B | 82.67 | #413 |
| 3 | InternVL3-78B | 80.26 | #361 |
| 4 | Qwen 2.5 VL 7B | 78.41 | #641 |
| 5 | Qwen 2 VL 7B | 76.42 | #743 |
| 6 | InternVL3-8B | 75.92 | #624 |
| 7 | Qwen2.5-Omni-7B | 75.07 | #633 |
| 8 | Claude 3.5 Sonnet | 73.22 | #362 |
| 9 | InternVL2-8B | 69.03 | #813 |
| 10 | Gemini 1.5 Flash | 63.57 | #620 |
| 11 | MiniCPM-V-2.6 | 62.5 | #827 |
| 12 | Gemini 2.0 Flash | 60.44 | #351 |
| 13 | Phi-4 Multimodal Instruct | 37.36 | #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-tempcompass-attribute-change · How It Works · Data refreshed daily, snapshot 2026-11-05.