# Human Preference Agreement on MM-RewardBench2 T2I Source: [Human Preference Agreement on MM-RewardBench2 T2I]() Collection index: [llms.txt]() Data updated: 2026-05-12T08:24:17.626Z Evaluation of model performance on the MM-RewardBench2 T2I dataset, measuring agreement with human preferences using 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 | Human Preference Agreement | | Dataset (including any stated split) | MM-RewardBench2 T2I | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 13 of 13 | | 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 3.1 Pro + ARR | 78.9 | Not recorded | 6a02b424dbc836b87d8246e4 | | Gemini 3.1 Pro | 75.1 | Not recorded | 6a02b424dbc836b87d8246e4 | | GPT-5 + ARR | 74.7 | Not recorded | 6a02b424dbc836b87d8246e4 | | GPT-5 | 70.5 | Not recorded | 6a02b424dbc836b87d8246e4 | | UnifiedReward-Thinking | 66 | Not recorded | 6a02b424dbc836b87d8246e4 | | Qwen3vl-8B + ARR | 62.7 | Not recorded | 6a02b424dbc836b87d8246e4 | | HPSv3 | 60.2 | Not recorded | 6a02b424dbc836b87d8246e4 | | UnifiedReward | 59.8 | Not recorded | 6a02b424dbc836b87d8246e4 | | PickScore | 58.6 | Not recorded | 6a02b424dbc836b87d8246e4 | | Qwen3-VL-8B | 57.6 | Not recorded | 6a02b424dbc836b87d8246e4 | | HPSv2 | 54.7 | Not recorded | 6a02b424dbc836b87d8246e4 | | ImageReward | 54 | Not recorded | 6a02b424dbc836b87d8246e4 | | CLIPScore | 51 | Not recorded | 6a02b424dbc836b87d8246e4 | ### Gemini 3.1 Pro + ARR Accuracy: 78.9. Result ID: 6a02e39f47f34ce924e59eea. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"ARR (Ours)","Evaluation Protocol":"ARR"}. This record does not assert independent reproduction. ### Gemini 3.1 Pro Accuracy: 75.1. Result ID: 6a02e39f47f34ce924e59ee7. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"VLM-as-Judge (Direct)","Evaluation Protocol":"Direct"}. This record does not assert independent reproduction. ### GPT-5 + ARR Accuracy: 74.7. Result ID: 6a02e39f47f34ce924e59ee9. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"ARR (Ours)","Evaluation Protocol":"ARR"}. This record does not assert independent reproduction. ### GPT-5 Accuracy: 70.5. Result ID: 6a02e39f47f34ce924e59ee6. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"VLM-as-Judge (Direct)","Evaluation Protocol":"Direct"}. This record does not assert independent reproduction. ### UnifiedReward-Thinking Accuracy: 66. Result ID: 6a02e39f47f34ce924e59ee2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Trained Reward Model"}. This record does not assert independent reproduction. ### Qwen3vl-8B + ARR Accuracy: 62.7. Result ID: 6a02e39f47f34ce924e59ee8. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"ARR (Ours)","Evaluation Protocol":"ARR"}. This record does not assert independent reproduction. ### HPSv3 Accuracy: 60.2. Result ID: 6a02e39f47f34ce924e59ee4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Trained Reward Model"}. This record does not assert independent reproduction. ### UnifiedReward Accuracy: 59.8. Result ID: 6a02e39f47f34ce924e59ee1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Trained Reward Model"}. This record does not assert independent reproduction. ### PickScore Accuracy: 58.6. Result ID: 6a02e39f47f34ce924e59edf. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"Trained Reward Model"}. This record does not assert independent reproduction. ### Qwen3-VL-8B Accuracy: 57.6. Result ID: 6a02e39f47f34ce924e59ee5. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"VLM-as-Judge (Direct)","Evaluation Protocol":"Direct"}. This record does not assert independent reproduction. ### HPSv2 Accuracy: 54.7. Result ID: 6a02e39f47f34ce924e59ee3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Trained Reward Model"}. This record does not assert independent reproduction. ### ImageReward Accuracy: 54. Result ID: 6a02e39f47f34ce924e59ee0. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"Trained Reward Model"}. This record does not assert independent reproduction. ### CLIPScore Accuracy: 51. Result ID: 6a02e39f47f34ce924e59ede. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Trained Reward Model"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.