# Long-form Video Question Answering on OPENTSUBENCH Source: [Long-form Video Question Answering on OPENTSUBENCH]() Collection index: [llms.txt]() Data updated: 2026-06-19T08:54:40.319Z Evaluation of model performance on the OPENTSUBENCH dataset for Long-form Video Question Answering, assessing capabilities across object centricity, temporal positioning, and compositional actions. 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 | Long-form Video Question Answering | | Dataset (including any stated split) | OPENTSUBENCH | | Metric | Object Centric 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 - [Object Centric Accuracy](): Higher is better; unit: PERCENT. - [Temporal Positioning Accuracy](): Higher is better; unit: PERCENT. - [Compositional Actions Accuracy](): Higher is better; unit: PERCENT. - [State Transition Accuracy](): Higher is better; unit: PERCENT. - [Long-Horizon Sparse Evidence Accuracy](): Higher is better; unit: PERCENT. - [Overall Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Object Centric Accuracy | Result date | Report paper | | --- | --- | --- | --- | | Qwen2.5-VL | 72.1 | Not recorded | 6a34cd54206542251d7c61e5 | | TimeSuite | 68.2 | Not recorded | 6a34cd54206542251d7c61e5 | | Videoagent | 65.1 | Not recorded | 6a34cd54206542251d7c61e5 | | VideoChat-Flash | 63.6 | Not recorded | 6a34cd54206542251d7c61e5 | | TimeChat | 62.8 | Not recorded | 6a34cd54206542251d7c61e5 | | VTimeLLM | 55.8 | Not recorded | 6a34cd54206542251d7c61e5 | | TIMEPROVE | 53.9 | Not recorded | 6a34cd54206542251d7c61e5 | | TIMEPROVE | 53.5 | Not recorded | 6a34cd54206542251d7c61e5 | | Time-R1 | 50.8 | Not recorded | 6a34cd54206542251d7c61e5 | | TIMEPROVE | 49.2 | Not recorded | 6a34cd54206542251d7c61e5 | | VideoTree | 36.8 | Not recorded | 6a34cd54206542251d7c61e5 | | GPT-4o | 27.9 | Not recorded | 6a34cd54206542251d7c61e5 | | AVP | 11.5 | Not recorded | 6a34cd54206542251d7c61e5 | | VideoLLaMA3 | 7.8 | Not recorded | 6a34cd54206542251d7c61e5 | ### Qwen2.5-VL Object Centric Accuracy: 72.1. Result ID: 6a3503c4fbc373592c328afc. [Reporting paper]() Recorded evaluation configuration: {"LLM":"Qwen 2.5"}. This record does not assert independent reproduction. ### TimeSuite Object Centric Accuracy: 68.2. Result ID: 6a3503c4fbc373592c328b01. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"LLM":"Mistral-7B"}. This record does not assert independent reproduction. ### Videoagent Object Centric Accuracy: 65.1. Result ID: 6a3503c4fbc373592c328b05. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"LLM":"GPT-4o","VLM":"GPT-4o"}. This record does not assert independent reproduction. ### VideoChat-Flash Object Centric Accuracy: 63.6. Result ID: 6a3503c4fbc373592c328afd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM":"InternLM2"}. This record does not assert independent reproduction. ### TimeChat Object Centric Accuracy: 62.8. Result ID: 6a3503c4fbc373592c328aff. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"LLM":"LLaMA-2-7B"}. This record does not assert independent reproduction. ### VTimeLLM Object Centric Accuracy: 55.8. Result ID: 6a3503c4fbc373592c328afe. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"LLM":"LLaMA-2-7B"}. This record does not assert independent reproduction. ### TIMEPROVE Object Centric Accuracy: 53.9. Result ID: 6a3503c4fbc373592c328b06. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM":"Gemma4-2B","VLM":"VLMA3"}. This record does not assert independent reproduction. ### TIMEPROVE Object Centric Accuracy: 53.5. Result ID: 6a3503c4fbc373592c328b07. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM":"Qwen2-7B","VLM":"VLMA3"}. This record does not assert independent reproduction. ### Time-R1 Object Centric Accuracy: 50.8. Result ID: 6a3503c4fbc373592c328b00. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM":"Qwen2.5VL"}. This record does not assert independent reproduction. ### TIMEPROVE Object Centric Accuracy: 49.2. Result ID: 6a3503c4fbc373592c328b08. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM":"Gemma4-2B","VLM":"GPT-4o"}. This record does not assert independent reproduction. ### VideoTree Object Centric Accuracy: 36.8. Result ID: 6a3503c4fbc373592c328b02. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"LLM":"GPT-5.1"}. This record does not assert independent reproduction. ### GPT-4o Object Centric Accuracy: 27.9. Result ID: 6a3503c4fbc373592c328b04. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM":"GPT-4o","VLM":"GPT-4o"}. This record does not assert independent reproduction. ### AVP Object Centric Accuracy: 11.5. Result ID: 6a3503c4fbc373592c328b03. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM":"GPT-4o","VLM":"GPT-4o"}. This record does not assert independent reproduction. ### VideoLLaMA3 Object Centric Accuracy: 7.8. Result ID: 6a3503c4fbc373592c328afb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"LLM":"Qwen 2"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.