# College-level Reasoning on MMMU (val) Source: [College-level Reasoning on MMMU (val)]() Collection index: [llms.txt]() Data updated: 2026-06-30T08:15:28.353Z Evaluation of model performance on the validation split of the MMMU dataset, focusing on College-level Reasoning capabilities. 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 | College-level Reasoning | | Dataset (including any stated split) | MMMU (val) | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 17 of 17 | | 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 | | --- | --- | --- | --- | | Qwen2.5VL-3b-Instruct | 53.1 | Not recorded | 6a434d7cba73165dbb743941 | | MoD-Multi Round | 49.47 | Not recorded | 6a434d7cba73165dbb743941 | | Self-Correction | 48.46 | Not recorded | 6a434d7cba73165dbb743941 | | Multi-Agent Debate | 48.28 | Not recorded | 6a434d7cba73165dbb743941 | | MoE-LoRA | 47.97 | Not recorded | 6a434d7cba73165dbb743941 | | Gemma-3-4B | 47.3 | Not recorded | 6a434d7cba73165dbb743941 | | MoD-Single Round | 46.44 | Not recorded | 6a434d7cba73165dbb743941 | | MoD-Multi Round | 38.44 | Not recorded | 6a434d7cba73165dbb743941 | | MoD-Single Round | 37.88 | Not recorded | 6a434d7cba73165dbb743941 | | MoE-LoRA | 37.29 | Not recorded | 6a434d7cba73165dbb743941 | | Qwen-VL-Chat | 37 | Not recorded | 6a434d7cba73165dbb743941 | | LLaVA-Next-Mistral-7B | 37 | Not recorded | 6a434d7cba73165dbb743941 | | ShareGPT4V-13B | 36.6 | Not recorded | 6a434d7cba73165dbb743941 | | Multi-Agent Debate | 36.22 | Not recorded | 6a434d7cba73165dbb743941 | | Self-Correction | 35.66 | Not recorded | 6a434d7cba73165dbb743941 | | LLaVA-v1.6-13b | 35 | Not recorded | 6a434d7cba73165dbb743941 | | InstructBLIP-7B | 30.6 | Not recorded | 6a434d7cba73165dbb743941 | ### Qwen2.5VL-3b-Instruct Accuracy: 53.1. Result ID: 6a437b06fbc373592c945ad1. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### MoD-Multi Round Accuracy: 49.47. Result ID: 6a437b06fbc373592c945ad6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5VL-3b-Instruct","Protocol":"multi-turn dialectical reasoning"}. This record does not assert independent reproduction. ### Self-Correction Accuracy: 48.46. Result ID: 6a437b06fbc373592c945ad2. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen2.5VL-3b-Instruct"}. This record does not assert independent reproduction. ### Multi-Agent Debate Accuracy: 48.28. Result ID: 6a437b06fbc373592c945ad3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5VL-3b-Instruct"}. This record does not assert independent reproduction. ### MoE-LoRA Accuracy: 47.97. Result ID: 6a437b06fbc373592c945ad4. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen2.5VL-3b-Instruct"}. This record does not assert independent reproduction. ### Gemma-3-4B Accuracy: 47.3. Result ID: 6a437b06fbc373592c945aca. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### MoD-Single Round Accuracy: 46.44. Result ID: 6a437b06fbc373592c945ad5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5VL-3b-Instruct","Protocol":"single-round inference"}. This record does not assert independent reproduction. ### MoD-Multi Round Accuracy: 38.44. Result ID: 6a437b06fbc373592c945ad0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"LLaVA-v1.6-13b","Protocol":"multi-turn dialectical reasoning"}. This record does not assert independent reproduction. ### MoD-Single Round Accuracy: 37.88. Result ID: 6a437b06fbc373592c945acf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"LLaVA-v1.6-13b","Protocol":"single-round inference"}. This record does not assert independent reproduction. ### MoE-LoRA Accuracy: 37.29. Result ID: 6a437b06fbc373592c945ace. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"LLaVA-v1.6-13b"}. This record does not assert independent reproduction. ### Qwen-VL-Chat Accuracy: 37. Result ID: 6a437b06fbc373592c945ac7. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### LLaVA-Next-Mistral-7B Accuracy: 37. Result ID: 6a437b06fbc373592c945ac9. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### ShareGPT4V-13B Accuracy: 36.6. Result ID: 6a437b06fbc373592c945ac8. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Multi-Agent Debate Accuracy: 36.22. Result ID: 6a437b06fbc373592c945acd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"LLaVA-v1.6-13b"}. This record does not assert independent reproduction. ### Self-Correction Accuracy: 35.66. Result ID: 6a437b06fbc373592c945acc. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"LLaVA-v1.6-13b"}. This record does not assert independent reproduction. ### LLaVA-v1.6-13b Accuracy: 35. Result ID: 6a437b06fbc373592c945acb. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### InstructBLIP-7B Accuracy: 30.6. Result ID: 6a437b06fbc373592c945ac6. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.