Multimodal benchmark
Video-MME leaderboard
As of October 2026, video-SALMONN 2+ has the highest published Video-MME score on Noometry at 79.7%, out of 15 models with results.
Last verified
About Video-MME
Multiple-choice questions about short, medium and long videos. Scores here are without subtitles.
- Category
- Multimodal
- Introduced
- 2024
- Size
- 900 videos
- Format
- Multiple choice
- Unit
- Percent (random guessing ≈ 25%)
- Official site
- video-mme.github.io
Top 15 models
- video-SALMONN 2+ 79.7%
- Gemini 1.5 Pro (May 2024) 75%
- Gemini 1.5 Pro (May 2024) 75%
- Qwen2.5-VL 72B Instruct 73.5%
- InternVL2_5-78B 72.1%
- GPT-4o 71.9%
- GPT-4o 71.9%
- Gemini 1.5 Flash (May 2024) 70.3%
- GPT-4o mini 64.8%
- NVILA 8B 64.2%
- InternVL2-40B 61.2%
- Claude 3.5 Sonnet 60%
- Claude 3.5 Sonnet 60%
- GPT-4V 59.9%
- Qwen-VL Max 51.3%
Sponsored placements are available on pages like this one. Advertise on Noometry
All results
| # | Model | Provider | Score | Setting | Source | Date |
|---|---|---|---|---|---|---|
| 1 | video-SALMONN 2+ |  ByteDance Seed | 79.7% | Epoch AI | ||
| 2 | Gemini 1.5 Pro (May 2024) |  Google | 75% | Epoch AI | ||
| 3 | Gemini 1.5 Pro (May 2024) | Google | 75% | Epoch AI | ||
| 4 | Qwen2.5-VL 72B Instruct |  Alibaba (Qwen) | 73.5% | Epoch AI | ||
| 5 | InternVL2_5-78B |  Shanghai AI Lab | 72.1% | 5-78B | Epoch AI | |
| 6 | GPT-4o | OpenAI | 71.9% | Epoch AI | ||
| 7 | GPT-4o | OpenAI | 71.9% | Epoch AI | ||
| 8 | Gemini 1.5 Flash (May 2024) | Google | 70.3% | Epoch AI | ||
| 9 | GPT-4o mini | OpenAI | 64.8% | Epoch AI | ||
| 10 | NVILA 8B |  NVIDIA | 64.2% | Epoch AI | ||
| 11 | InternVL2-40B | Shanghai AI Lab | 61.2% | Epoch AI | ||
| 12 | Claude 3.5 Sonnet | Anthropic | 60% | Epoch AI | ||
| 13 | Claude 3.5 Sonnet | Anthropic | 60% | Epoch AI | ||
| 14 | GPT-4V | OpenAI | 59.9% | Epoch AI | ||
| 15 | Qwen-VL Max | Alibaba (Qwen) | 51.3% | Epoch AI |
Compare the leaders
Other multimodal benchmarks
- LMArena Vision
- GeoBench
- VPCT
- Blueprint-Bench 2
- Furniture Assembly
- LMArena Document (reference)
- MindCube (reference)
- ScienceQA (reference)
- SpatialViz-Bench (reference)
Frequently asked questions
What does Video-MME measure?
Multiple-choice questions about short, medium and long videos. Scores here are without subtitles.
Which model has the highest Video-MME score?
As of October 2026, video-SALMONN 2+ has the highest published Video-MME score on Noometry at 79.7%, out of 15 models with results.
What is the best open-weight model on Video-MME?
Qwen2.5-VL 72B Instruct has the highest Video-MME accuracy among open-weight models at 73.5%, ranking 4 of 15 overall.