# Fine-grained audio-visual video understanding on OmniVideoBench Source: [Fine-grained audio-visual video understanding on OmniVideoBench]() Collection index: [llms.txt]() Data updated: 2026-06-23T07:53:25.356Z Evaluation of fine-grained audio-visual video understanding capabilities on the OmniVideoBench dataset, measured by accuracy. 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 | Fine-grained audio-visual video understanding | | Dataset (including any stated split) | OmniVideoBench | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 14 of 14 | | 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. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | Gemini-2.5-Pro | 58.9 | Not recorded | 6a0fe328f984739ed6739ea8 | | Gemini-2.0-Flash | 41.5 | Not recorded | 6a0fe328f984739ed6739ea8 | | CogniRoute | 38.9 | Not recorded | 6a3a131a206542251d7c7027 | | Qwen3 Omni 30B | 38.4 | Not recorded | 6a3a131a206542251d7c7027 | | LatentOmni | 35.4 | Not recorded | 6a0fe328f984739ed6739ea8 | | Qwen2.5-Omni-7B + Explicit Text CoT | 33.2 | Not recorded | 6a0fe328f984739ed6739ea8 | | OmniVinci | 32.1 | Not recorded | 6a0fe328f984739ed6739ea8 | | Baichuan-Omni-1.5 | 30.7 | Not recorded | 6a0fe328f984739ed6739ea8 | | VITA-1.5-7B | 30.5 | Not recorded | 6a0fe328f984739ed6739ea8 | | HumanOmniV2-7B | 30.5 | Not recorded | 6a0fe328f984739ed6739ea8 | | Qwen2.5-Omni-7B + Vanilla SFT | 30.5 | Not recorded | 6a0fe328f984739ed6739ea8 | | MiniCPM-o-7B | 29.7 | Not recorded | 6a0fe328f984739ed6739ea8 | | Qwen2.5-Omni-7B | 29.3 | Not recorded | 6a0fe328f984739ed6739ea8 | | VideoLLaMA2-7B | 29.2 | Not recorded | 6a0fe328f984739ed6739ea8 | ### Gemini-2.5-Pro Accuracy: 58.9. Result ID: 6a100e71fbc373592c08daeb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Accessibility":"Proprietary"}. This record does not assert independent reproduction. ### Gemini-2.0-Flash Accuracy: 41.5. Result ID: 6a100e71fbc373592c08daea. [Reporting paper]() Recorded evaluation configuration: {"Model Accessibility":"Proprietary"}. This record does not assert independent reproduction. ### CogniRoute Accuracy: 38.9. Result ID: 6a3a3b6cfbc373592c343ab4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Qwen3 Omni 30B Accuracy: 38.4. Result ID: 6a3a3b6cfbc373592c343ab3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Parameters":"30B"}. This record does not assert independent reproduction. ### LatentOmni Accuracy: 35.4. Result ID: 6a100e71fbc373592c08dae9. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Accessibility":"Open-source","Training/Reasoning Strategy":"Latent reasoning"}. This record does not assert independent reproduction. ### Qwen2.5-Omni-7B + Explicit Text CoT Accuracy: 33.2. Result ID: 6a100e71fbc373592c08dae7. [Reporting paper]() Recorded evaluation configuration: {"Model Accessibility":"Open-source","Training/Reasoning Strategy":"Explicit Text CoT"}. This record does not assert independent reproduction. ### OmniVinci Accuracy: 32.1. Result ID: 6a100e71fbc373592c08dae5. [Reporting paper]() Recorded evaluation configuration: {"Model Accessibility":"Open-source"}. This record does not assert independent reproduction. ### Baichuan-Omni-1.5 Accuracy: 30.7. Result ID: 6a100e71fbc373592c08dae4. [Reporting paper]() Recorded evaluation configuration: {"Model Accessibility":"Open-source"}. This record does not assert independent reproduction. ### VITA-1.5-7B Accuracy: 30.5. Result ID: 6a100e71fbc373592c08dae2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Accessibility":"Open-source"}. This record does not assert independent reproduction. ### HumanOmniV2-7B Accuracy: 30.5. Result ID: 6a100e71fbc373592c08dae3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Accessibility":"Open-source"}. This record does not assert independent reproduction. ### Qwen2.5-Omni-7B + Vanilla SFT Accuracy: 30.5. Result ID: 6a100e71fbc373592c08dae8. [Reporting paper]() Recorded evaluation configuration: {"Model Accessibility":"Open-source","Training/Reasoning Strategy":"Vanilla SFT"}. This record does not assert independent reproduction. ### MiniCPM-o-7B Accuracy: 29.7. Result ID: 6a100e71fbc373592c08dae1. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Accessibility":"Open-source"}. This record does not assert independent reproduction. ### Qwen2.5-Omni-7B Accuracy: 29.3. Result ID: 6a100e71fbc373592c08dae6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Accessibility":"Open-source","Training/Reasoning Strategy":"Base"}. This record does not assert independent reproduction. ### VideoLLaMA2-7B Accuracy: 29.2. Result ID: 6a100e71fbc373592c08dae0. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Accessibility":"Open-source"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.