Multimodal Understanding on MMMU (Score)
81.8MMMU ScoreGPT-5-Thinking
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5-ThinkingModel Category=Close-source Models2025.09 | 81.8 | |
| Gemini 2.5 FlashSize=-, mode=instruct2026.04 | 76.3 | |
| Gemini-2.5-ProModel Category=Close-source Models2025.09 | 74.7 | |
| InternVL3.5Size=8B, mode=instruct2026.04 | 73.4 | |
| Audio Flamingo3Model Category=Specialized Models2026.04 | 72.42 | |
| Ming-OmniModel Category=Unified Models2026.04 | 70.8 | |
| LongCat-Next2026.03 | 70.6 | |
| Qwen3-VLSize=8B, mode=instruct2026.04 | 69.6 | |
| Qwen3-OmniSize=30B-A3B, mode=instruct2026.04 | 69.1 | |
| NEO-unify2026.03 | 68.9 | |
| MiniCPM-o 4.5Size=9B, mode=instruct2026.04 | 67.6 | |
| SPOT-EBase Model=Qwen3-VL-32B2026.06 | 64.6 | |
| Qwen2.5-Omni-3BModel Category=Specialized Models2026.04 | 63.3 | |
| Audio-OmniModel Category=Unified Models2026.04 | 63.3 | |
| Qwen3-VL-32BBase Model=Qwen3-VL-32B2026.06 | 63.2 | |
| SPOT-EBase Model=Qwen2.5-VL-32B2026.06 | 63 | |
| SPOT-EBase Model=InternVL2.5-26B2026.06 | 62.7 | |
| Qwen2.5-VL-32BBase Model=Qwen2.5-VL-32B2026.06 | 61.4 | |
| InternVL2.5-26BBase Model=InternVL2.5-26B2026.06 | 60.8 | |
| VAPO-Thinker-7BModel Category=Our Models2025.09 | 60.2 | |
| VLAA-Thinker-7BModel Category=Open-source Models2025.09 | 59.1 | |
| Lumina2026.03 | 58.6 | |
| UniWorld-V12026.03 | 58.6 | |
| UniWorld-V1Params=12B2025.11 | 58.6 | |
| Qwen2.5-VLLLM=Qwen2.5-7B2025.05 | 58.6 | |
| Vision-R1-7BModel Category=Open-source Models2025.09 | 57.6 | |
| Qwen2-Audio-InstructModel Category=Specialized Models2026.04 | 57.4 | |
| InternVL2.5-8BModel Category=Open-source Models2025.09 | 56.2 | |
| InternVL2.5LLM=InternLM2.5-7B2025.05 | 56 | |
| VAPO-Thinker-3BModel Category=Our Models2025.09 | 55.6 | |
| BAGEL2026.03 | 55.3 | |
| BAGELParams=14B2025.11 | 55.3 | |
| InternVL-U2026.03 | 54.7 | |
| R1-OneVision-7BModel Category=Open-source Models2025.09 | 54.3 | |
| Bagel+LatentUMM2026.05 | 53.2 | |
| OmniGen22026.03 | 53.1 | |
| OminiGen2Params=7B2025.11 | 53.1 | |
| Bagel+UniCot2026.05 | 53.1 | |
| Qwen2.5-VL-7BModel Category=Open-source Models2025.09 | 52.7 | |
| Bagel+UniGame2026.05 | 52.4 | |
| Bagel+RecA2026.05 | 52.3 | |
| BagelParams=14B2026.05 | 51.9 | |
| Bagel+SFT2026.05 | 51.9 | |
| SPOT-EBase Model=LLaVA-OV-72B2026.06 | 51.8 | |
| Ovis-U12026.03 | 51.1 | |
| BLIP3-o2026.03 | 50.6 | |
| BLIP3-oParams=8B2025.11 | 50.6 | |
| LLaVA-OV-72BBase Model=LLaVA-OV-72B2026.06 | 49.8 | |
| Show-o22026.03 | 48.9 | |
| LLaVA-OneVisionLLM=Qwen2-7B2025.05 | 48.8 | |
| Show-O2Params=7B2026.05 | 47.9 | |
| Slot-MLLM (14B)LLM=Qwen2.5-14B-Instruct2025.05 | 46.7 | |
| Omnigen2Params=7B2026.05 | 46 | |
| Mogao2026.03 | 44.2 | |
| UniGameParams=7B2025.11 | 43.8 | |
| TokenFlow-XLParams=14B2025.11 | 43.2 | |
| OneCAT2026.03 | 41.9 | |
| Qwen2-VL-2BLLM=Qwen2-1.5B, Retain Ratio (%)=100%2026.05 | 41.7 | |
| SFTParams=7B2025.11 | 41.2 | |
| Janus-Pro2026.03 | 41 | |
| Janus-Pro†Params=7B2025.11 | 41 | |
| Janus ProParams=7B2026.05 | 40.7 | |
| VEN-VLLLM=Qwen3-1.7B, Retain Ratio (%)=∼ 7.5%2026.05 | 39.7 | |
| MiniCPM-V2LLM=MiniCPM-2.4B, Retain Ratio (%)=100%2026.05 | 39.6 | |
| TokenFlow-XLType=Und. & Gen., # Params=13B2025.09 | 38.7 | |
| ILLUMEType=Und. & Gen., # Params=7B2025.09 | 38.2 | |
| VEN-VLLLM=Qwen2.5-1.5B, Retain Ratio (%)=∼ 10%2026.05 | 37.9 | |
| Show-O2Params=1.5B2026.05 | 37.1 | |
| TinyLLaVA†LLM=Qwen2.5-1.5B, Retain Ratio (%)=100%2026.05 | 37 | |
| Janus-Pro-1BType=Und. & Gen., # Params=1.5B2025.09 | 36.3 | |
| VisionZipLLM=Vicuna-7B, Retain Ratio (%)=∼ 10%2026.05 | 36.2 | |
| Qwen-VL-ChatLLM=Qwen-7B, Retain Ratio (%)=100%2026.05 | 35.9 | |
| LLaVA-NeXTLLM=Vicuna-7B2025.05 | 35.8 | |
| RecAParams=1.5B2025.11 | 35.7 | |
| SEED-XParams=7B2025.11 | 35.6 | |
| VEN-VLLLM=Qwen3-0.6B, Retain Ratio (%)=∼ 7.5%2026.05 | 35.6 | |
| LLaVA-v1.5Type=Und. Only, # Params=7B2025.09 | 35.4 | |
| OursType=Und. & Gen., # Params=1.5B2025.09 | 35.4 | |
| PinPointTraining Data=GQA, Avg. FLOPs(T)=24.39 (61.1%)2026.03 | 35 | |
| InternVL2-2BLLM=InternLM2-1.8B, Retain Ratio (%)=100%2026.05 | 34.9 | |
| LLaVA-1.5LLM=Vicuna-7B, Retain Ratio (%)=100%2026.05 | 34.4 | |
| FUDOKIType=Und. & Gen., # Params=1.5B2025.09 | 34.3 | |
| PDropTraining Data=Train-free, Avg. FLOPs(T)=26.00 (65.1%)2026.03 | 34.11 | |
| VanillaTraining Data=-, Avg. FLOPs(T)=39.92 (100.0%)2026.03 | 34.1 | |
| TinyLLaVA†LLM=Qwen1.5-1.8B, Retain Ratio (%)=100%2026.05 | 34.1 | |
| FastVLLM=Vicuna-7B, Retain Ratio (%)=∼ 10%2026.05 | 34 | |
| TinyLLaVA†LLM=Qwen2.5-0.5B, Retain Ratio (%)=100%2026.05 | 33.6 | |
| FastVTraining Data=Train-free, Avg. FLOPs(T)=28.17 (70.6%)2026.03 | 33.56 | |
| Slot-MLLM (7B)LLM=Vicuna-7B2025.05 | 33.2 | |
| SparseVLMLLM=Vicuna-7B, Retain Ratio (%)=∼ 10%2026.05 | 32.7 | |
| Mini-Gemini-2BLLM=Gemma-2B, Retain Ratio (%)=100%2026.05 | 31.7 | |
| Emu3Params=8B2025.11 | 31.6 | |
| Emu3-ChatType=Und. Only, # Params=8B2025.09 | 31.6 | |
| TinyLLaVA†LLM=Qwen1.5-0.5B, Retain Ratio (%)=100%2026.05 | 31.6 | |
| Emu3Params=8B2026.05 | 31.4 | |
| LLaVA-Phi-1.5Type=Und. Only, # Params=1.3B2025.09 | 30.7 | |
| JanusType=Und. & Gen., # Params=1.5B2025.09 | 30.5 | |
| Emu3.5Params=34B2026.05 | 29.2 | |
| Janus FlowParams=1.3B2026.05 | 29 | |
| MMaDAParams=8B2026.05 | 28.9 |