Multimodal Reasoning on MMMU (Accuracy)
72.9AccuracyGPT-4o
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-4oModel Category=Proprietary Models2025.06 | 72.9 | |
| Gemini-2.0-ProModel Category=Proprietary Models2025.06 | 72.6 | |
| Claude-3.7-SonnetModel Category=Proprietary Models2025.06 | 71 | |
| InternVL2.5-78BModel Category=Open-source Models (>70B)2025.06 | 70.1 | |
| InternVL2.5-78B-MPOModel Category=Open-source Models (>70B)2025.06 | 68.2 | |
| Qwen2.5-VL-72BModel Category=Open-source Models (>70B)2025.06 | 68.2 | |
| PRCRModel=InternVL3.5, #Params=8B2026.06 | 66.14 | |
| Token-ReplayModel=InternVL3.5, #Params=8B2026.06 | 65.82 | |
| BaselineModel=InternVL3.5, #Params=8B2026.06 | 65.05 | |
| PRCRModel=Qwen3-VL, #Params=32B2026.06 | 64.89 | |
| Token-ReplayModel=Qwen3-VL, #Params=32B2026.06 | 64.73 | |
| BaselineModel=Qwen3-VL, #Params=32B2026.06 | 63.68 | |
| PRCRModel=InternVL3.5, #Params=14B2026.06 | 63.2 | |
| Token-ReplayModel=InternVL3.5, #Params=14B2026.06 | 62.88 | |
| BaselineModel=InternVL3.5, #Params=14B2026.06 | 62.02 | |
| SVIPPRM Trainable Params=7B2026.06 | 61.3 | |
| Athena-7BModel Category=Open-source Models (~7B)2025.06 | 61.1 | |
| Visual-PRMPRM Trainable Params=7B2026.06 | 60.2 | |
| PRCRModel=Qwen3-VL, #Params=8B2026.06 | 60.07 | |
| Token-ReplayModel=Qwen3-VL, #Params=8B2026.06 | 59.45 | |
| STRIDEBase model=Qwen2.5-VL-7B2026.06 | 58.7 | |
| Qwen2.5-VL-7BModel Category=Open-source Models (~7B)2025.06 | 58 | |
| SRPOBase model=Qwen2.5-VL-7B2026.06 | 57.1 | |
| VL-PRM300KPRM Trainable Params=8B2026.06 | 56.5 | |
| InternVL2.5-8BModel Category=Open-source Models (~7B)2025.06 | 56.2 | |
| BaselineModel=Qwen3-VL, #Params=8B2026.06 | 55.22 | |
| MM-EurekaBase model=Qwen2.5-VL-7B2026.06 | 55.2 | |
| MMS-PRMPRM Trainable Params=0M2026.06 | 54.2 | |
| BaseBase model=Qwen2.5-VL-7B2026.06 | 54.1 | |
| Op-SkipBackbone=Qwen3-VL-8B, TFLOPs=66%, N=202026.06 | 52.89 | |
| APETBackbone=Qwen3-VL-8B, TFLOPs=66%2026.06 | 52.53 | |
| V2DropBackbone=Qwen3-VL-8B, TFLOPs=66%2026.06 | 52.44 | |
| ShortVBackbone=Qwen3-VL-8B, TFLOPs=58%, N=202026.06 | 52.22 | |
| AdaSteerModel=QWEN3.52026.06 | 52 | |
| MARSModel=INTERNVL 3.52026.06 | 51.4 | |
| VanillaBackbone=Qwen3-VL-8B, TFLOPs=100%2026.06 | 51.33 | |
| VSkipBackbone=Qwen3-VL-8B, TFLOPs=76%, N=202026.06 | 51.33 | |
| Zero-shotModel=INTERNVL 3.52026.06 | 51.3 | |
| ECSOModel=INTERNVL 3.52026.06 | 51.3 | |
| MiniCPM-o-2.6-8BModel Category=Open-source Models (~7B)2025.06 | 50.9 | |
| AdaSteerModel=INTERNVL 3.52026.06 | 50.9 | |
| Zero-shotModel=QWEN3-VL2026.06 | 50.8 | |
| ECSOModel=QWEN3-VL2026.06 | 50.8 | |
| MARSModel=QWEN3-VL2026.06 | 50.8 | |
| Zero-shotModel=QWEN3.52026.06 | 50.8 | |
| ECSOModel=QWEN3.52026.06 | 50.8 | |
| AdaSteerModel=QWEN3-VL2026.06 | 50.4 | |
| AdaSteerModel=MOLMO22026.06 | 50.2 | |
| APETBackbone=Qwen2.5-VL-7B, TFLOPs=55%2026.06 | 50.11 | |
| Zero-shotModel=MOLMO22026.06 | 50.1 | |
| ECSOModel=MOLMO22026.06 | 50.1 | |
| MARSModel=MOLMO22026.06 | 50.1 | |
| VanillaBackbone=Qwen2.5-VL-7B, TFLOPs=100%2026.06 | 50 | |
| V2DropBackbone=Qwen2.5-VL-7B, TFLOPs=55%2026.06 | 49.89 | |
| SASAModel=QWEN3-VL2026.06 | 49.2 | |
| SASAModel=MOLMO22026.06 | 48.2 | |
| VTWBackbone=Qwen3-VL-8B, TFLOPs=58%, K=162026.06 | 47.33 | |
| VSkipBackbone=Qwen2.5-VL-7B, TFLOPs=67%, N=162026.06 | 47 | |
| Op-SkipBackbone=Qwen2.5-VL-7B, TFLOPs=55%, N=202026.06 | 46.89 | |
| VTWBackbone=Qwen2.5-VL-7B, TFLOPs=50%, K=162026.06 | 43.56 | |
| ShortVBackbone=Qwen2.5-VL-7B, TFLOPs=50%, N=202026.06 | 41.89 | |
| SASAModel=INTERNVL 3.52026.06 | 41.3 | |
| Zero-shotModel=GEMMA32026.06 | 39.8 | |
| ECSOModel=GEMMA32026.06 | 39.8 | |
| AdaSteerModel=GEMMA32026.06 | 39.8 | |
| MARSModel=GEMMA32026.06 | 39.8 | |
| SASAModel=GEMMA32026.06 | 39.4 | |
| MARSModel=QWEN3.52026.06 | 38.2 | |
| VSkip+Backbone=LLaVA-1.5-7B, TFLOPs=76%, N=202026.06 | 36.44 | |
| VanillaBackbone=LLaVA-1.5-7B, TFLOPs=100%2026.06 | 36.11 | |
| APETBackbone=LLaVA-1.5-7B, TFLOPs=66%2026.06 | 36.11 | |
| VTWBackbone=LLaVA-1.5-7B, TFLOPs=55%, K=162026.06 | 35.67 | |
| ShortVBackbone=LLaVA-1.5-7B, TFLOPs=55%, N=202026.06 | 35.67 | |
| Op-SkipBackbone=LLaVA-1.5-7B, TFLOPs=66%, N=202026.06 | 35.67 | |
| VSkipBackbone=LLaVA-1.5-7B, TFLOPs=76%, N=202026.06 | 35.56 | |
| V2DropBackbone=LLaVA-1.5-7B, TFLOPs=66%2026.06 | 31.67 | |
| SASAModel=QWEN3.52026.06 | 30.8 |