# Video Understanding on MLVU (Score, Avg Score, Acc. %) Source: [Video Understanding on MLVU (Score, Avg Score, Acc. %)]() Collection index: [llms.txt]() Data updated: 2026-07-02T09:35:57.620Z Evaluation of video understanding models on the MLVU dataset, reporting performance across Score, Average Score, and Accuracy metrics. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Video Understanding | | Dataset (including any stated split) | MLVU | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 42 of 42 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Accuracy](): Higher is better; unit: PERCENT. - [Score](): Higher is better; unit: SCALAR. - [Average Score](): Higher is better; unit: SCALAR. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | SCORE | 102.8 | Not recorded | 69ca03720636e5bf4afc92c2 | | SCORE | 102.4 | Not recorded | 69ca03720636e5bf4afc92c2 | | VidCom^2 | 102.1 | Not recorded | 69ca03720636e5bf4afc92c2 | | DyCoke | 101.7 | Not recorded | 69ca03720636e5bf4afc92c2 | | VidCom^2 | 101.6 | Not recorded | 69ca03720636e5bf4afc92c2 | | FastVID | 100.3 | Not recorded | 69ca03720636e5bf4afc92c2 | | Vanilla | 100 | Not recorded | 69ca03720636e5bf4afc92c2 | | FastVID | 99.8 | Not recorded | 69ca03720636e5bf4afc92c2 | | SCORE | 99.5 | Not recorded | 69ca03720636e5bf4afc92c2 | | HoliTom | 99.3 | Not recorded | 69ca03720636e5bf4afc92c2 | | HoliTom | 97.1 | Not recorded | 69ca03720636e5bf4afc92c2 | | FastVID | 95.9 | Not recorded | 69ca03720636e5bf4afc92c2 | | VidCom^2 | 95.3 | Not recorded | 69ca03720636e5bf4afc92c2 | | DyCoke | 94.8 | Not recorded | 69ca03720636e5bf4afc92c2 | | HoliTom | 92.3 | Not recorded | 69ca03720636e5bf4afc92c2 | | QCA | 77.1 | Not recorded | 6a45f095dc71372d7795a361 | | InternVL3.5-38B | 77 | Not recorded | 6a45f095dc71372d7795a361 | | QCA | 75.7 | Not recorded | 6a45f095dc71372d7795a361 | | Qwen2.5-VL-72B | 74.6 | Not recorded | 6a45f095dc71372d7795a361 | | LLaVA-Video-72B | 74.4 | Not recorded | 6a45f095dc71372d7795a361 | | Kimi-VL-16B-A3B | 74.2 | Not recorded | 6a45f095dc71372d7795a361 | | Qwen3-VL-30B-A3B | 72.8 | Not recorded | 6a45f095dc71372d7795a361 | | Qwen3-VL-8B | 71 | Not recorded | 6a45f095dc71372d7795a361 | | NVILA | 70.1 | Not recorded | 6a45f095dc71372d7795a361 | | mPLUG-Owl3 | 70 | Not recorded | 6a45f095dc71372d7795a361 | | ProtoKV | 68.9 | Not recorded | 6a3e07c1206542251d7cb69a | | Apollo | 68.7 | Not recorded | 6a45f095dc71372d7795a361 | | InfiniPot-V | 68.4 | Not recorded | 6a3e07c1206542251d7cb69a | | InfiniPot-V | 67.2 | Not recorded | 6a3e07c1206542251d7cb69a | | ProtoKV | 67 | Not recorded | 6a3e07c1206542251d7cb69a | | GPT-4o | 64.6 | Not recorded | 6a45f095dc71372d7795a361 | | DyCoke | 60.5 | Not recorded | 6a3e07c1206542251d7cb69a | | DyCoke | 57.5 | Not recorded | 6a3e07c1206542251d7cb69a | ### SCORE Accuracy: 102.8. Result ID: 69ca289c47f34ce924b27d22. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"25%","Max Frames":"128"}. This record does not assert independent reproduction. ### SCORE Accuracy: 102.4. Result ID: 69ca289c47f34ce924b27d1d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"40%","Max Frames":"128"}. This record does not assert independent reproduction. ### VidCom^2 Accuracy: 102.1. Result ID: 69ca289c47f34ce924b27d1b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"40%","Max Frames":"128"}. This record does not assert independent reproduction. ### DyCoke Accuracy: 101.7. Result ID: 69ca289c47f34ce924b27d19. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Retention Ratio R":"40%","Max Frames":"128"}. This record does not assert independent reproduction. ### VidCom^2 Accuracy: 101.6. Result ID: 69ca289c47f34ce924b27d20. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"25%","Max Frames":"128"}. This record does not assert independent reproduction. ### FastVID Accuracy: 100.3. Result ID: 69ca289c47f34ce924b27d1c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"40%","Max Frames":"128"}. This record does not assert independent reproduction. ### Vanilla Accuracy: 100. Result ID: 69ca289c47f34ce924b27d18. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"100%","Max Frames":"128"}. This record does not assert independent reproduction. ### FastVID Accuracy: 99.8. Result ID: 69ca289c47f34ce924b27d21. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"25%","Max Frames":"128"}. This record does not assert independent reproduction. ### SCORE Accuracy: 99.5. Result ID: 69ca289c47f34ce924b27d26. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"10%","Max Frames":"128"}. This record does not assert independent reproduction. ### HoliTom Accuracy: 99.3. Result ID: 69ca289c47f34ce924b27d1a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"40%","Max Frames":"128"}. This record does not assert independent reproduction. ### HoliTom Accuracy: 97.1. Result ID: 69ca289c47f34ce924b27d1f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"25%","Max Frames":"128"}. This record does not assert independent reproduction. ### FastVID Accuracy: 95.9. Result ID: 69ca289c47f34ce924b27d25. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"10%","Max Frames":"128"}. This record does not assert independent reproduction. ### VidCom^2 Accuracy: 95.3. Result ID: 69ca289c47f34ce924b27d24. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"10%","Max Frames":"128"}. This record does not assert independent reproduction. ### DyCoke Accuracy: 94.8. Result ID: 69ca289c47f34ce924b27d1e. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Retention Ratio R":"25%","Max Frames":"128"}. This record does not assert independent reproduction. ### HoliTom Accuracy: 92.3. Result ID: 69ca289c47f34ce924b27d23. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Retention Ratio R":"10%","Max Frames":"128"}. This record does not assert independent reproduction. ### QCA Accuracy: 77.1. Result ID: 6a4630f960b88bb339863164. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"30B","Frames":"64","Backbone":"Qwen3-VL-30B-A3B"}. This record does not assert independent reproduction. ### InternVL3.5-38B Accuracy: 77. Result ID: 6a4630f960b88bb33986315d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"38B","Frames":"64"}. This record does not assert independent reproduction. ### QCA Accuracy: 75.7. Result ID: 6a4630f960b88bb339863162. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"8B","Frames":"64","Backbone":"Qwen3-VL-8B"}. This record does not assert independent reproduction. ### Qwen2.5-VL-72B Accuracy: 74.6. Result ID: 6a4630f960b88bb33986315a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"72B","Frames":"1fps"}. This record does not assert independent reproduction. ### LLaVA-Video-72B Accuracy: 74.4. Result ID: 6a4630f960b88bb33986315c. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"LLM Size":"72B","Frames":"64"}. This record does not assert independent reproduction. ### Kimi-VL-16B-A3B Accuracy: 74.2. Result ID: 6a4630f960b88bb33986315b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"16B","Frames":"64"}. This record does not assert independent reproduction. ### Qwen3-VL-30B-A3B Accuracy: 72.8. Result ID: 6a4630f960b88bb339863163. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"30B","Frames":"64"}. This record does not assert independent reproduction. ### Qwen3-VL-8B Accuracy: 71. Result ID: 6a4630f960b88bb339863161. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"8B","Frames":"64"}. This record does not assert independent reproduction. ### NVILA Accuracy: 70.1. Result ID: 6a4630f960b88bb33986315e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"7B","Frames":"256"}. This record does not assert independent reproduction. ### mPLUG-Owl3 Accuracy: 70. Result ID: 6a4630f960b88bb33986315f. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"LLM Size":"8B","Frames":"128"}. This record does not assert independent reproduction. ### ProtoKV Accuracy: 68.9. Result ID: 6a3e35c1fbc373592c6008b7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Budget \|M\|":"6K","Backbone":"Qwen2-VL-7B"}. This record does not assert independent reproduction. ### Apollo Accuracy: 68.7. Result ID: 6a4630f960b88bb339863160. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"7B","Frames":"2fps"}. This record does not assert independent reproduction. ### InfiniPot-V Accuracy: 68.4. Result ID: 6a3e35c1fbc373592c6008b5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Budget \|M\|":"6K","Backbone":"Qwen2-VL-7B"}. This record does not assert independent reproduction. ### InfiniPot-V Accuracy: 67.2. Result ID: 6a3e35c1fbc373592c6008b4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Budget \|M\|":"3K","Backbone":"Qwen2-VL-7B"}. This record does not assert independent reproduction. ### ProtoKV Accuracy: 67. Result ID: 6a3e35c1fbc373592c6008b6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Budget \|M\|":"3K","Backbone":"Qwen2-VL-7B"}. This record does not assert independent reproduction. ### GPT-4o Accuracy: 64.6. Result ID: 6a4630f960b88bb339863159. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM Size":"-","Frames":"256 / 0.5fps"}. This record does not assert independent reproduction. ### DyCoke Accuracy: 60.5. Result ID: 6a3e35c1fbc373592c6008b3. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Budget \|M\|":"6K","Backbone":"Qwen2-VL-7B"}. This record does not assert independent reproduction. ### DyCoke Accuracy: 57.5. Result ID: 6a3e35c1fbc373592c6008b2. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Budget \|M\|":"3K","Backbone":"Qwen2-VL-7B"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.