Video-MME (with subtitles)
Multimodal · 2024-05-31
Video-MME overall multiple-choice accuracy on its 2,700 questions over 900 videos when the model receives the sampled frames plus the video's subtitles, without audio. It is a component of the Video-MME family, whose primary is the no-subtitle condition (videomme); the audio-visual-subtitle condition is the separate diagnostic videomme-audio-visual-subtitles. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| Qwen3.8-Max | 90.4 |
| Kimi K3 | 90.0 |
| GPT-5.6 Sol | 89.5 |
| GPT-5.4 | 89.5 |
| GPT-5.5 | 89.3 |
| Gemini 3 Pro Preview | 88.4 |
| Qwen3.7-Plus (source unspecified) | 88.0 |
| Qwen3.6 Plus (2026-04-02) | 87.8 |
| Qwen3.6 27B | 87.7 |
| Qwen3.5 397B A17B | 87.5 |
| Kimi K2.5 | 87.4 |
| Qwen3.5 122B A10B | 87.3 |
| Qwen3.5-27B | 87.0 |
| Qwen3.5 35B A3B | 86.6 |
| Qwen3.6 35B A3B | 86.6 |
| Opus 4.6 | 86.1 |
| GPT-5.2 | 86.0 |
| Opus 4.8 | 86.0 |
| Qwen3.5-9B | 84.5 |
| Qwen3.5-4B | 83.5 |
| GPT-5 Mini | 83.5 |
| GPT-5.6 Luna | 81.6 |
| Sonnet 4.5 | 81.1 |
| Qwen2.5-VL-72B-Instruct | 79.1 |
| Qwen3 VL 235B A22B Thinking | 79.0 |
| Qwen2.5 VL 32B Instruct | 77.9 |
| Gemini 3.5 Flash-Lite | 77.8 |
| Qwen2 VL 72B Instruct | 77.8 |
| Opus 4.5 | 77.6 |
| Gemini 2.5 Flash-Lite | 74.6 |
| InternVL2.5-78B | 74.0 |
| Gemma 4 12B IT | 73.3 |
| Kimi-VL-A3B-Thinking-2506 | 71.9 |
| GPT-5 Nano | 71.7 |
| Qwen2.5-VL-7B-Instruct | 71.6 |
| LFM3-VL-2B | 69.0 |
| GPT-4o Mini | 68.9 |
| Kimi-VL-A3B-Thinking | 66.0 |
| Qwen3.5-0.8B | 63.8 |
| MiniCPM-V 2.6 | 63.6 |