# Multimodal mathematical reasoning on MathVista (test) Source: [Multimodal mathematical reasoning on MathVista (test)]() Collection index: [llms.txt]() Data updated: 2026-03-31T08:40:59.377Z Evaluation of model performance on the test split of the MathVista dataset for Multimodal Mathematical Reasoning, reporting Accuracy and response length statistics. 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 | Multimodal mathematical reasoning | | Dataset (including any stated split) | MathVista (test) | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 34 of 34 | | 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. - [Avg Response Length (tokens)](): Higher is better; unit: SCALAR. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | Qwen-VL-Max | 74.2 | Not recorded | 69608b6253fdd5c361744094 | | GPRO-7B | 74.2 | Not recorded | 69608b6253fdd5c361744094 | | FAST-7B | 73.8 | Not recorded | 69608b6253fdd5c361744094 | | Vision-R1 | 73.5 | Not recorded | 69608b6253fdd5c361744094 | | GSPO + KAWHI (Qwen3) | 72.1 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | GRPO + KAWHI (Qwen3) | 71.8 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | DAPO + KAWHI (Qwen3) | 71.21 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | GRPO (Qwen3) | 71.07 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | MM-R1 | 71 | Not recorded | 69608b6253fdd5c361744094 | | GSPO (Qwen3) | 70.5 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | Qwen3-VL-4B-Instruct Base | 70.43 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | OpenVLThinker | 70.2 | Not recorded | 69608b6253fdd5c361744094 | | GSPO | 69.42 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | DAPO (Qwen3) | 69.4 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | GSPO + KAWHI | 69.37 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | GRPO + KAWHI | 69.3 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | DAPO | 69.2 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | DAPO + KAWHI | 69.08 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | VPPO | 68.42 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | GRPO | 68.37 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | Qwen2.5-VL-7B | 68.2 | Not recorded | 69608b6253fdd5c361744094 | | Claude-3.5 Sonnet | 67.7 | Not recorded | 69608b6253fdd5c361744094 | | FT-RL | 67.68 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | Qwen2.5-VL-7B-Instruct Base | 67.47 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | Step-GRPO | 66.87 | Not recorded | 69cb54fe0636e5bf4afc9e86 | | GPRO-3B | 66.8 | Not recorded | 69608b6253fdd5c361744094 | | FAST-3B | 66.2 | Not recorded | 69608b6253fdd5c361744094 | | R1-OneVision | 64.1 | Not recorded | 69608b6253fdd5c361744094 | | GPT-4o | 63.8 | Not recorded | 69608b6253fdd5c361744094 | | LMM-R1 | 63.2 | Not recorded | 69608b6253fdd5c361744094 | | Qwen2.5-VL-3B | 62.3 | Not recorded | 69608b6253fdd5c361744094 | | Mulberry | 62.1 | Not recorded | 69608b6253fdd5c361744094 | | Curr-ReFT | 61.9 | Not recorded | 69608b6253fdd5c361744094 | | Qwen2-VL-7B | 58.2 | Not recorded | 69608b6253fdd5c361744094 | ### Qwen-VL-Max Accuracy: 74.2. Result ID: 6997dafa5698665ec4334d7b. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"Closed-Source Models"}. This record does not assert independent reproduction. ### GPRO-7B Accuracy: 74.2. Result ID: 6997dafa5698665ec4334d89. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"7B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### FAST-7B Accuracy: 73.8. Result ID: 6997dafa5698665ec4334d88. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"7B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### Vision-R1 Accuracy: 73.5. Result ID: 6997dafa5698665ec4334d85. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"7B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### GSPO + KAWHI (Qwen3) Accuracy: 72.1. Result ID: 69cb888a47f34ce924b3dfb6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen3-VL-4B-Instruct","Reward Strategy":"uniform reward","RL Method":"KAWHI"}. This record does not assert independent reproduction. ### GRPO + KAWHI (Qwen3) Accuracy: 71.8. Result ID: 69cb888a47f34ce924b3dfb2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen3-VL-4B-Instruct","Reward Strategy":"uniform reward","RL Method":"KAWHI"}. This record does not assert independent reproduction. ### DAPO + KAWHI (Qwen3) Accuracy: 71.21. Result ID: 69cb888a47f34ce924b3dfb4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen3-VL-4B-Instruct","Reward Strategy":"uniform reward","RL Method":"KAWHI"}. This record does not assert independent reproduction. ### GRPO (Qwen3) Accuracy: 71.07. Result ID: 69cb888a47f34ce924b3dfb1. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone Model":"Qwen3-VL-4B-Instruct","Reward Strategy":"uniform reward","RL Method":"GRPO"}. This record does not assert independent reproduction. ### MM-R1 Accuracy: 71. Result ID: 6997dafa5698665ec4334d84. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"7B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### GSPO (Qwen3) Accuracy: 70.5. Result ID: 69cb888a47f34ce924b3dfb5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen3-VL-4B-Instruct","Reward Strategy":"uniform reward","RL Method":"GSPO"}. This record does not assert independent reproduction. ### Qwen3-VL-4B-Instruct Base Accuracy: 70.43. Result ID: 69cb888a47f34ce924b3dfb0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen3-VL-4B-Instruct","Reward Strategy":"none","RL Method":"Base"}. This record does not assert independent reproduction. ### OpenVLThinker Accuracy: 70.2. Result ID: 6997dafa5698665ec4334d87. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"7B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### GSPO Accuracy: 69.42. Result ID: 69cb888a47f34ce924b3dfae. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"uniform reward","RL Method":"GSPO"}. This record does not assert independent reproduction. ### DAPO (Qwen3) Accuracy: 69.4. Result ID: 69cb888a47f34ce924b3dfb3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen3-VL-4B-Instruct","Reward Strategy":"uniform reward","RL Method":"DAPO"}. This record does not assert independent reproduction. ### GSPO + KAWHI Accuracy: 69.37. Result ID: 69cb888a47f34ce924b3dfaf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"uniform reward","RL Method":"KAWHI"}. This record does not assert independent reproduction. ### GRPO + KAWHI Accuracy: 69.3. Result ID: 69cb888a47f34ce924b3dfab. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"uniform reward","RL Method":"KAWHI"}. This record does not assert independent reproduction. ### DAPO Accuracy: 69.2. Result ID: 69cb888a47f34ce924b3dfac. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"uniform reward","RL Method":"DAPO"}. This record does not assert independent reproduction. ### DAPO + KAWHI Accuracy: 69.08. Result ID: 69cb888a47f34ce924b3dfad. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"uniform reward","RL Method":"KAWHI"}. This record does not assert independent reproduction. ### VPPO Accuracy: 68.42. Result ID: 69cb888a47f34ce924b3dfa9. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"fine-grained reward","RL Method":"VPPO"}. This record does not assert independent reproduction. ### GRPO Accuracy: 68.37. Result ID: 69cb888a47f34ce924b3dfaa. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"uniform reward","RL Method":"GRPO"}. This record does not assert independent reproduction. ### Qwen2.5-VL-7B Accuracy: 68.2. Result ID: 6997dafa5698665ec4334d83. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"7B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### Claude-3.5 Sonnet Accuracy: 67.7. Result ID: 6997dafa5698665ec4334d7a. [Reporting paper]() Recorded evaluation configuration: {"Model Category":"Closed-Source Models"}. This record does not assert independent reproduction. ### FT-RL Accuracy: 67.68. Result ID: 69cb888a47f34ce924b3dfa8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"fine-grained reward","RL Method":"FT-RL"}. This record does not assert independent reproduction. ### Qwen2.5-VL-7B-Instruct Base Accuracy: 67.47. Result ID: 69cb888a47f34ce924b3dfa6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"none","RL Method":"Base"}. This record does not assert independent reproduction. ### Step-GRPO Accuracy: 66.87. Result ID: 69cb888a47f34ce924b3dfa7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone Model":"Qwen2.5-VL-7B-Instruct","Reward Strategy":"fine-grained reward","RL Method":"Step-GRPO"}. This record does not assert independent reproduction. ### GPRO-3B Accuracy: 66.8. Result ID: 6997dafa5698665ec4334d82. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"3B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### FAST-3B Accuracy: 66.2. Result ID: 6997dafa5698665ec4334d81. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"3B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### R1-OneVision Accuracy: 64.1. Result ID: 6997dafa5698665ec4334d86. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"7B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### GPT-4o Accuracy: 63.8. Result ID: 6997dafa5698665ec4334d79. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Closed-Source Models"}. This record does not assert independent reproduction. ### LMM-R1 Accuracy: 63.2. Result ID: 6997dafa5698665ec4334d80. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"3B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### Qwen2.5-VL-3B Accuracy: 62.3. Result ID: 6997dafa5698665ec4334d7e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"3B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### Mulberry Accuracy: 62.1. Result ID: 6997dafa5698665ec4334d7d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"7B","Base Architecture":"Qwen2-VL"}. This record does not assert independent reproduction. ### Curr-ReFT Accuracy: 61.9. Result ID: 6997dafa5698665ec4334d7f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Base Model Scale":"3B","Base Architecture":"Qwen2.5-VL"}. This record does not assert independent reproduction. ### Qwen2-VL-7B Accuracy: 58.2. Result ID: 6997dafa5698665ec4334d7c. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Base Model Scale":"7B","Base Architecture":"Qwen2-VL"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.