Video-MME
Video understanding across durations, without subtitles.
As of 19 Sept 2026, video-SALMONN 2+ leads Video-MME on BenchLeader with 79.7%, ahead of Gemini 1.5 Pro 001 at 75.0%, across 42 model configurations with a published result.
- Published by
- Video-MMEdata via Epoch AI Benchmarking Hub
- Category
- Multimodal
- Index weight
- Reference only
- Models
- 42
- Data as of
- 19 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
What the test looks like
Questions about videos from seconds to an hour long, answered without subtitles.
How it is scored
Accuracy, as published by Video-MME.
What to keep in mind
Video-capable models only.
- 1video-SALMONN 2+79.7%
- 2Gemini 1.5 Pro 00175.0%
- 3Gemini 1.5 Pro 001 Feb2475.0%
- 4Qwen2.5-VL 72B Instruct73.5%
- 5InternVL2_5-78B72.1%
- 6GPT-4o71.9%
- 7Qwen2 Vl 72B71.2%
- 8LLaVA-Video-72B-Qwen270.6%
- 9Gemini 1.5 Flash 00170.3%
- 10ViLAMP-llava-qwen67.5%
- 11Oryx-1.5-32B67.3%
- 12LLaVA-OneVision 72B66.3%
- 13VideoLLAMA3-7B66.2%
- 14LLaVA-Video-7B-Qwen265.9%
- 15LLaVA-Video-7B-Qwen2-TPO65.6%
42 of 42
| # | ||||
|---|---|---|---|---|
| 1 |  video-SALMONN 2+ByteDance | 79.7% | – | 2025-06-18 |
| 2 |  Gemini 1.5 Pro 001Google | 75.0% | 45.1 | 2024-05-14 |
| 3 | Gemini 1.5 Pro 001 Feb24Google | 75.0% | – | 2024-02-15 |
| 4 |  Qwen2.5-VL 72B InstructAlibabaopen ↗ | 73.5% | – | 2024-09-19 |
| 5 |  InternVL2_5-78BShanghai AI Lab | 72.1% | – | 2024-12-06 |
| 6 |  GPT-4oOpenAI | 71.9% | 44.3 | 2024-11-20 |
| 7 | Qwen2 Vl 72BAlibaba | 71.2% | – | 2024-08-29 |
| 8 |  LLaVA-Video-72B-Qwen2OpenBMB | 70.6% | – | 2024-09-02 |
| 9 | Gemini 1.5 Flash 001Google | 70.3% | 36.7 | 2024-05-23 |
| 10 | ViLAMP-llava-qwenUnknown | 67.5% | – | 2025-05-01 |
| 11 | Oryx-1.5-32BUnknown | 67.3% | – | 2024-10-22 |
| 12 | LLaVA-OneVision 72BOpenBMB | 66.3% | – | 2024-08-06 |
| 13 | VideoLLAMA3-7BUnknown | 66.2% | – | 2024-01-22 |
| 14 | LLaVA-Video-7B-Qwen2OpenBMB | 65.9% | – | 2024-09-02 |
| 15 | LLaVA-Video-7B-Qwen2-TPOOpenBMB | 65.6% | – | 2025-01-19 |
| 16 | VideoChat-Flash-Qwen2-7B_res448Unknown | 65.3% | – | 2025-01-11 |
| 17 | GPT-4o miniOpenAI | 64.8% | 36.2 | 2024-07-18 |
| 18 | ByteVideoLLM-14BUnknown | 64.6% | – | 2024-10-13 |
| 19 |  NVILA 8BNVIDIA | 64.2% | – | 2024-12-10 |
| 20 | LiveCC-7B-InstructUnknown | 64.1% | – | 2024-04-12 |
| 21 | Qwen2 Vl 7BAlibaba | 63.9% | – | 2024-08-29 |
| 22 | MiniCPM-o-2_6OpenBMB | 63.9% | – | 2025-01-12 |
| 23 | VideoLLAMA2-7BUnknown | 62.4% | – | 2024-06-12 |
| 24 | InternVL2-40BShanghai AI Lab | 61.2% | – | 2024-07-08 |
| 25 | Minicpm V 2.6OpenBMB | 60.9% | – | 2024-08-03 |
| 26 |  Claude 3.5 SonnetAnthropic | 60.0% | 46.6 | 2024-06-20 |
| 27 | GPT-4VOpenAI | 59.9% | – | 2023-11-06 |
| 28 | mPLUG-Owl3-7B-241101Unknown | 59.3% | – | 2024-11-26 |
| 29 | VITA-1.5Unknown | 56.1% | – | 2024-12-20 |
| 30 | Video-XL-7BUnknown | 55.5% | – | 2024-10-17 |
| 31 | Video-CCAM-7B-v1.2Unknown | 53.2% | – | 2024-09-29 |
| 32 | long-llava-qwen2-7bUnknown | 52.9% | – | 2024-08-30 |
| 33 | LongVA-7BUnknown | 52.6% | – | 2024-06-13 |
| 34 | Qwen VlmaxAlibaba | 51.3% | – | 2024-01-18 |
| 35 | SliME-Llama3-8BUnknown | 45.3% | – | 2024-06-02 |
| 36 | Chat-Uni-Vi-7B-v1.5 + 100k SG-WVUnknown | 43.2% | – | 2024-06-20 |
| 37 | Qwen VlAlibaba | 41.1% | – | 2023-08-20 |
| 38 | Chat-UniVi-7B-v1.5Unknown | 40.6% | – | 2024-04-23 |
| 39 | sharegpt4video-8bUnknown | 39.9% | – | 2024-05-27 |
| 40 | Video-LLaVA-7BUnknown | 39.9% | – | 2023-11-17 |
| 41 |  video_chat2_mistralMistral AI | 39.5% | – | 2023-11-29 |
| 42 | ST-LLMUnknown | 37.9% | – | 2024-03-28 |
Cite as: BenchLeader, “Video-MME leaderboard”, https://www.benchleader.com/benchmarks/video_mme, data as of 19 Sept 2026.
Video-MME: questions
- What does Video-MME measure?
- Questions about videos from seconds to an hour long, answered without subtitles. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads Video-MME?
- video-SALMONN 2+ leads Video-MME with 79.7% as of 19 Sept 2026, ahead of Gemini 1.5 Pro 001 at 75.0%.
- How many models have Video-MME results?
- 42 model configurations have a Video-MME result on BenchLeader, all taken from Video-MME via Epoch AI Benchmarking Hub.
- Who runs Video-MME and how often is it updated?
- Video-MME is published by Video-MME. BenchLeader re-reads the published results every morning and records the date each result was published.
- Does Video-MME count toward the BenchLeader Index?
- No. Video-MME is shown for reference but left out of the composite index.