# General Multimodal Understanding on MMMU Pro Source: [General Multimodal Understanding on MMMU Pro]() Collection index: [llms.txt]() Data updated: 2026-06-09T09:37:49.288Z Evaluation of model performance on the MMMU Pro dataset for General Multimodal Understanding capabilities, 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 | General Multimodal Understanding | | Dataset (including any stated split) | MMMU Pro | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 31 of 31 | | 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 | | --- | --- | --- | --- | | Qwen3-VL-32B-Instruct | 56.9 | Not recorded | 6a152902dbd7150506f7ddd9 | | IVR-R1 | 53.3 | Not recorded | 6a152902dbd7150506f7ddd9 | | Vision-R1 | 52.5 | Not recorded | 6a152902dbd7150506f7ddd9 | | Vision-SR1 | 49.3 | Not recorded | 6a152902dbd7150506f7ddd9 | | Supervised Fine-tuning (before RL) | 48.6 | Not recorded | 6a152902dbd7150506f7ddd9 | | IVR-R1 | 47.8 | Not recorded | 6a152902dbd7150506f7ddd9 | | Vision-R1 | 47.2 | Not recorded | 6a152902dbd7150506f7ddd9 | | Vision-SR1 | 45.5 | Not recorded | 6a152902dbd7150506f7ddd9 | | Supervised Fine-tuning (before RL) | 43.6 | Not recorded | 6a152902dbd7150506f7ddd9 | | Zero-shot Inference (before RL) | 42.7 | Not recorded | 6a152902dbd7150506f7ddd9 | | VPPO | 38.8 | Not recorded | 6a279e6e90966ade919bfac6 | | PRPO | 38.8 | Not recorded | 6a279e6e90966ade919bfac6 | | VL-Rethinker-7B | 37.1 | Not recorded | 6a279e6e90966ade919bfac6 | | Qwen2.5-VL-72B-Instruct | 36.6 | Not recorded | 6a152902dbd7150506f7ddd9 | | PAPO-D | 36.3 | Not recorded | 6a279e6e90966ade919bfac6 | | PAPO-G | 35.5 | Not recorded | 6a279e6e90966ade919bfac6 | | GRPO | 35.2 | Not recorded | 6a279e6e90966ade919bfac6 | | R1-ShareVL-7B | 35.1 | Not recorded | 6a279e6e90966ade919bfac6 | | NoisyRollout-7B | 34.5 | Not recorded | 6a279e6e90966ade919bfac6 | | PRPO | 30.3 | Not recorded | 6a279e6e90966ade919bfac6 | | MM-Eureka-7B | 30.3 | Not recorded | 6a279e6e90966ade919bfac6 | | DAPO | 29 | Not recorded | 6a279e6e90966ade919bfac6 | | PAPO-D | 28.8 | Not recorded | 6a279e6e90966ade919bfac6 | | VPPO | 28.8 | Not recorded | 6a279e6e90966ade919bfac6 | | DAPO | 28.4 | Not recorded | 6a279e6e90966ade919bfac6 | | PAPO-G | 26.4 | Not recorded | 6a279e6e90966ade919bfac6 | | GRPO | 25.7 | Not recorded | 6a279e6e90966ade919bfac6 | | Qwen2.5-VL-7B | 25.2 | Not recorded | 6a279e6e90966ade919bfac6 | | Qwen2.5-VL-32B-Instruct | 21.7 | Not recorded | 6a152902dbd7150506f7ddd9 | | Qwen2.5-VL-3B | 19.3 | Not recorded | 6a279e6e90966ade919bfac6 | | Zero-shot Inference (before RL) | 16.2 | Not recorded | 6a152902dbd7150506f7ddd9 | ### Qwen3-VL-32B-Instruct Accuracy: 56.9. Result ID: 6a154936fbc373592c0c38ef. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone model":"Qwen3-VL-32B-Instruct"}. This record does not assert independent reproduction. ### IVR-R1 Accuracy: 53.3. Result ID: 6a154936fbc373592c0c38f9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone model":"Qwen3-VL-4B"}. This record does not assert independent reproduction. ### Vision-R1 Accuracy: 52.5. Result ID: 6a154936fbc373592c0c38f7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone model":"Qwen3-VL-4B"}. This record does not assert independent reproduction. ### Vision-SR1 Accuracy: 49.3. Result ID: 6a154936fbc373592c0c38f8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone model":"Qwen3-VL-4B"}. This record does not assert independent reproduction. ### Supervised Fine-tuning (before RL) Accuracy: 48.6. Result ID: 6a154936fbc373592c0c38f6. [Reporting paper]() Recorded evaluation configuration: {"Backbone model":"Qwen3-VL-4B"}. This record does not assert independent reproduction. ### IVR-R1 Accuracy: 47.8. Result ID: 6a154936fbc373592c0c38f4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone model":"Qwen2.5-VL-7B"}. This record does not assert independent reproduction. ### Vision-R1 Accuracy: 47.2. Result ID: 6a154936fbc373592c0c38f2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone model":"Qwen2.5-VL-7B"}. This record does not assert independent reproduction. ### Vision-SR1 Accuracy: 45.5. Result ID: 6a154936fbc373592c0c38f3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone model":"Qwen2.5-VL-7B"}. This record does not assert independent reproduction. ### Supervised Fine-tuning (before RL) Accuracy: 43.6. Result ID: 6a154936fbc373592c0c38f1. [Reporting paper]() Recorded evaluation configuration: {"Backbone model":"Qwen2.5-VL-7B"}. This record does not assert independent reproduction. ### Zero-shot Inference (before RL) Accuracy: 42.7. Result ID: 6a154936fbc373592c0c38f5. [Reporting paper]() Recorded evaluation configuration: {"Backbone model":"Qwen3-VL-4B"}. This record does not assert independent reproduction. ### VPPO Accuracy: 38.8. Result ID: 6a27deeafbc373592c2507ab. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Scale":"7B","Base Model":"Qwen2.5-VL-7B","Rollout":"8"}. This record does not assert independent reproduction. ### PRPO Accuracy: 38.8. Result ID: 6a27deeafbc373592c2507ac. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"7B","Base Model":"Qwen2.5-VL-7B","Rollout":"8"}. This record does not assert independent reproduction. ### VL-Rethinker-7B Accuracy: 37.1. Result ID: 6a27deeafbc373592c2507a3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"7B","Rollout":"8"}. This record does not assert independent reproduction. ### Qwen2.5-VL-72B-Instruct Accuracy: 36.6. Result ID: 6a154936fbc373592c0c38ed. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone model":"Qwen2.5-VL-72B-Instruct"}. This record does not assert independent reproduction. ### PAPO-D Accuracy: 36.3. Result ID: 6a27deeafbc373592c2507aa. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"7B","Base Model":"Qwen2.5-VL-7B","Rollout":"8"}. This record does not assert independent reproduction. ### PAPO-G Accuracy: 35.5. Result ID: 6a27deeafbc373592c2507a9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"7B","Base Model":"Qwen2.5-VL-7B","Rollout":"8"}. This record does not assert independent reproduction. ### GRPO Accuracy: 35.2. Result ID: 6a27deeafbc373592c2507a7. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Scale":"7B","Base Model":"Qwen2.5-VL-7B","Rollout":"8"}. This record does not assert independent reproduction. ### R1-ShareVL-7B Accuracy: 35.1. Result ID: 6a27deeafbc373592c2507a5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"7B","Rollout":"8"}. This record does not assert independent reproduction. ### NoisyRollout-7B Accuracy: 34.5. Result ID: 6a27deeafbc373592c2507a4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"7B","Rollout":"8"}. This record does not assert independent reproduction. ### PRPO Accuracy: 30.3. Result ID: 6a27deeafbc373592c2507a1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"3B","Base Model":"Qwen2.5-VL-3B","Rollout":"8"}. This record does not assert independent reproduction. ### MM-Eureka-7B Accuracy: 30.3. Result ID: 6a27deeafbc373592c2507a2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"7B","Rollout":"8"}. This record does not assert independent reproduction. ### DAPO Accuracy: 29. Result ID: 6a27deeafbc373592c2507a8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"7B","Base Model":"Qwen2.5-VL-7B","Rollout":"8"}. This record does not assert independent reproduction. ### PAPO-D Accuracy: 28.8. Result ID: 6a27deeafbc373592c25079f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"3B","Base Model":"Qwen2.5-VL-3B","Rollout":"8"}. This record does not assert independent reproduction. ### VPPO Accuracy: 28.8. Result ID: 6a27deeafbc373592c2507a0. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Scale":"3B","Base Model":"Qwen2.5-VL-3B","Rollout":"8"}. This record does not assert independent reproduction. ### DAPO Accuracy: 28.4. Result ID: 6a27deeafbc373592c25079d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"3B","Base Model":"Qwen2.5-VL-3B","Rollout":"8"}. This record does not assert independent reproduction. ### PAPO-G Accuracy: 26.4. Result ID: 6a27deeafbc373592c25079e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Scale":"3B","Base Model":"Qwen2.5-VL-3B","Rollout":"8"}. This record does not assert independent reproduction. ### GRPO Accuracy: 25.7. Result ID: 6a27deeafbc373592c25079c. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Scale":"3B","Base Model":"Qwen2.5-VL-3B","Rollout":"8"}. This record does not assert independent reproduction. ### Qwen2.5-VL-7B Accuracy: 25.2. Result ID: 6a27deeafbc373592c2507a6. [Reporting paper]() Recorded evaluation configuration: {"Model Scale":"7B","Rollout":"8"}. This record does not assert independent reproduction. ### Qwen2.5-VL-32B-Instruct Accuracy: 21.7. Result ID: 6a154936fbc373592c0c38ee. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone model":"Qwen2.5-VL-32B-Instruct"}. This record does not assert independent reproduction. ### Qwen2.5-VL-3B Accuracy: 19.3. Result ID: 6a27deeafbc373592c25079b. [Reporting paper]() Recorded evaluation configuration: {"Model Scale":"3B","Rollout":"8"}. This record does not assert independent reproduction. ### Zero-shot Inference (before RL) Accuracy: 16.2. Result ID: 6a154936fbc373592c0c38f0. [Reporting paper]() Recorded evaluation configuration: {"Backbone model":"Qwen2.5-VL-7B"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.