# Medical Visual Question Answering on MMMU H&M Source: [Medical Visual Question Answering on MMMU H&M]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:31:38.680Z Evaluation of model performance on the MMMU H&M (Health & Medicine) dataset subset for medical visual question answering capabilities, reporting overall 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 | Medical Visual Question Answering | | Dataset (including any stated split) | MMMU H&M | | Metric | Accuracy | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 25 of 25 | | 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: SCALAR. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | GPT-o4-mini | 0.7875 | Not recorded | 6965d1af6343fa78c83238cb | | GPT-5 | 0.7786 | Not recorded | 6965d1af6343fa78c83238cb | | Claude-4.5-sonnet | 0.7673 | Not recorded | 6965d1af6343fa78c83238cb | | GPT-5-mini | 0.7428 | Not recorded | 6965d1af6343fa78c83238cb | | Claude-4.5-haiku | 0.7214 | Not recorded | 6965d1af6343fa78c83238cb | | Gemini-2.5-pro | 0.7129 | Not recorded | 6965d1af6343fa78c83238cb | | MEDVISTA-R1 | 0.5643 | Not recorded | 6965d1af6343fa78c83238cb | | InternVL3-8B | 0.5589 | Not recorded | 6965d1af6343fa78c83238cb | | Cold-start w/o Tools | 0.55 | Not recorded | 6965d1af6343fa78c83238cb | | Qwen3vl-8B | 0.5286 | Not recorded | 6965d1af6343fa78c83238cb | | Cold-start w/o Reasoning | 0.5214 | Not recorded | 6965d1af6343fa78c83238cb | | MEDVISTAGYM (InternVL3-8B) | 0.4857 | Not recorded | 6965d1af6343fa78c83238cb | | MEDVISTA-R1 | 0.4743 | Not recorded | 6965d1af6343fa78c83238cb | | Gemini-2.5-flash | 0.4643 | Not recorded | 6965d1af6343fa78c83238cb | | Qwen2.5vl-7B | 0.4643 | Not recorded | 6965d1af6343fa78c83238cb | | Cold-start w/o Reasoning | 0.4571 | Not recorded | 6965d1af6343fa78c83238cb | | GRPO w/o Tools | 0.453 | Not recorded | 6965d1af6343fa78c83238cb | | Cold-start w/o Tools | 0.4429 | Not recorded | 6965d1af6343fa78c83238cb | | Direct GRPO w/o cold-start | 0.438 | Not recorded | 6965d1af6343fa78c83238cb | | Internvl3-2B | 0.4357 | Not recorded | 6965d1af6343fa78c83238cb | | MEDVISTAGYM (Qwen3vl-8B) | 0.4286 | Not recorded | 6965d1af6343fa78c83238cb | | Direct GRPO w/o cold-start | 0.42 | Not recorded | 6965d1af6343fa78c83238cb | | LLaVA-Med-7B | 0.3771 | Not recorded | 6965d1af6343fa78c83238cb | | GRPO w/o Tools | 0.375 | Not recorded | 6965d1af6343fa78c83238cb | | MEDVISTAGYM (InternVL3-2B) | 0.3643 | Not recorded | 6965d1af6343fa78c83238cb | ### GPT-o4-mini Accuracy: 0.7875. Result ID: 6997cc685698665ec423a198. [Reporting paper]() Recorded evaluation configuration: {"Model size":"Proprietary","Prompting protocol":"vanilla prompt"}. This record does not assert independent reproduction. ### GPT-5 Accuracy: 0.7786. Result ID: 6997cc685698665ec423a196. [Reporting paper]() Recorded evaluation configuration: {"Model size":"Proprietary","Prompting protocol":"vanilla prompt"}. This record does not assert independent reproduction. ### Claude-4.5-sonnet Accuracy: 0.7673. Result ID: 6997cc685698665ec423a19a. [Reporting paper]() Recorded evaluation configuration: {"Model size":"Proprietary","Prompting protocol":"vanilla prompt"}. This record does not assert independent reproduction. ### GPT-5-mini Accuracy: 0.7428. Result ID: 6997cc685698665ec423a197. [Reporting paper]() Recorded evaluation configuration: {"Model size":"Proprietary","Prompting protocol":"vanilla prompt"}. This record does not assert independent reproduction. ### Claude-4.5-haiku Accuracy: 0.7214. Result ID: 6997cc685698665ec423a199. [Reporting paper]() Recorded evaluation configuration: {"Model size":"Proprietary","Prompting protocol":"vanilla prompt"}. This record does not assert independent reproduction. ### Gemini-2.5-pro Accuracy: 0.7129. Result ID: 6997cc685698665ec423a19b. [Reporting paper]() Recorded evaluation configuration: {"Model size":"Proprietary","Prompting protocol":"vanilla prompt"}. This record does not assert independent reproduction. ### MEDVISTA-R1 Accuracy: 0.5643. Result ID: 6997cc685698665ec423a1a7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters"}. This record does not assert independent reproduction. ### InternVL3-8B Accuracy: 0.5589. Result ID: 6997cc685698665ec423a1a1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters"}. This record does not assert independent reproduction. ### Cold-start w/o Tools Accuracy: 0.55. Result ID: 6997cc685698665ec423a1a5. [Reporting paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters","Tool-access setting":"removed","SFT-stage setting":"Cold-start"}. This record does not assert independent reproduction. ### Qwen3vl-8B Accuracy: 0.5286. Result ID: 6997cc685698665ec423a19f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters"}. This record does not assert independent reproduction. ### Cold-start w/o Reasoning Accuracy: 0.5214. Result ID: 6997cc685698665ec423a1a6. [Reporting paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters","Reasoning-stage setting":"removed","SFT-stage setting":"Cold-start"}. This record does not assert independent reproduction. ### MEDVISTAGYM (InternVL3-8B) Accuracy: 0.4857. Result ID: 6997cc685698665ec423a1a2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters","Tool-access setting":"enabled"}. This record does not assert independent reproduction. ### MEDVISTA-R1 Accuracy: 0.4743. Result ID: 6997cc685698665ec423a1ae. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model size":"\< 7B parameters"}. This record does not assert independent reproduction. ### Gemini-2.5-flash Accuracy: 0.4643. Result ID: 6997cc685698665ec423a19c. [Reporting paper]() Recorded evaluation configuration: {"Model size":"Proprietary","Prompting protocol":"vanilla prompt"}. This record does not assert independent reproduction. ### Qwen2.5vl-7B Accuracy: 0.4643. Result ID: 6997cc685698665ec423a19d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters"}. This record does not assert independent reproduction. ### Cold-start w/o Reasoning Accuracy: 0.4571. Result ID: 6997cc685698665ec423a1ad. [Reporting paper]() Recorded evaluation configuration: {"Model size":"\< 7B parameters","Reasoning-stage setting":"removed","SFT-stage setting":"Cold-start"}. This record does not assert independent reproduction. ### GRPO w/o Tools Accuracy: 0.453. Result ID: 6997cc685698665ec423a1a3. [Reporting paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters","Tool-access setting":"removed","Reasoning-stage setting":"GRPO"}. This record does not assert independent reproduction. ### Cold-start w/o Tools Accuracy: 0.4429. Result ID: 6997cc685698665ec423a1ac. [Reporting paper]() Recorded evaluation configuration: {"Model size":"\< 7B parameters","Tool-access setting":"removed","SFT-stage setting":"Cold-start"}. This record does not assert independent reproduction. ### Direct GRPO w/o cold-start Accuracy: 0.438. Result ID: 6997cc685698665ec423a1a4. [Reporting paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters","SFT-stage setting":"removed","Reasoning-stage setting":"Direct GRPO"}. This record does not assert independent reproduction. ### Internvl3-2B Accuracy: 0.4357. Result ID: 6997cc685698665ec423a1a8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model size":"\< 7B parameters"}. This record does not assert independent reproduction. ### MEDVISTAGYM (Qwen3vl-8B) Accuracy: 0.4286. Result ID: 6997cc685698665ec423a1a0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model size":"7-13B parameters","Tool-access setting":"enabled"}. This record does not assert independent reproduction. ### Direct GRPO w/o cold-start Accuracy: 0.42. Result ID: 6997cc685698665ec423a1ab. [Reporting paper]() Recorded evaluation configuration: {"Model size":"\< 7B parameters","SFT-stage setting":"removed","Reasoning-stage setting":"Direct GRPO"}. This record does not assert independent reproduction. ### LLaVA-Med-7B Accuracy: 0.3771. Result ID: 6997cc685698665ec423a19e. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model size":"7-13B parameters"}. This record does not assert independent reproduction. ### GRPO w/o Tools Accuracy: 0.375. Result ID: 6997cc685698665ec423a1aa. [Reporting paper]() Recorded evaluation configuration: {"Model size":"\< 7B parameters","Tool-access setting":"removed","Reasoning-stage setting":"GRPO"}. This record does not assert independent reproduction. ### MEDVISTAGYM (InternVL3-2B) Accuracy: 0.3643. Result ID: 6997cc685698665ec423a1a9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model size":"\< 7B parameters","Tool-access setting":"enabled"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.