Multimodal Understanding on MM-Vet
82.2MM-Vet ScoreInternVL3.5-38B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| InternVL3.5-38BLLM=Qwen3-32B2025.12 | 82.2 | — | — | |
| InternVL3Params=8B, Types=AR, Architecture Category=Und. Only2025.11 | 81.3 | — | — | |
| Qwen3-VL + DiGLLM=Qwen3-4B2025.12 | 77.3 | — | — | |
| Qwen3-VL + DiGLLM=Qwen3-8B2025.12 | 76.8 | — | — | |
| Qwen3-VLLLM=Qwen3-8B2025.12 | 76.4 | — | — | |
| Qwen2.5-VL 72BLLM=Qwen2.5-72B2025.12 | 76.2 | — | — | |
| Qwen3-VLLLM=Qwen3-4B2025.12 | 76 | — | — | |
| VLSI-7BSize=7B2024.12 | 75.8 | — | — | |
| GPT-4o (0806)Model Version=08062024.12 | 75.1 | — | — | |
| Qwen2-VL-72BSize=72B2024.12 | 73.9 | — | — | |
| EMMAParams=4B2025.12 | 73 | — | — | |
| Ming-Lite-UniParams=8B+1.6B, Architecture category=Multi-modal Understanding Models as Enhanced Conditional Encoders2026.03 | 72.3 | — | — | |
| MM1#param=30B, open-source=false2024.04 | 72.1 | — | — | |
| VLsIParameters=7B2024.12 | 70.8 | — | — | |
| Phantom-7BSize=7B2024.12 | 70.8 | — | — | |
| Qwen2.5VL 7B# Vis tok.=14002025.04 | 69.7 | — | — | |
| BAGEL (Ours)Baseline=BAGEL, Params=7B active2026.06 | 69.5 | — | — | |
| ResDecBase Model=Qwen2.5-VL2026.02 | 68.7 | — | — | |
| BLIP3o-8B (Ours)Baseline=BLIP3o-8B, Params=8B2026.06 | 68.7 | — | — | |
| VISTABase Model=Qwen2.5-VL2026.02 | 67.8 | — | — | |
| GPT-4V2024.01 | 67.7 | — | — | |
| GPT-4Vopen-source=false2024.04 | 67.6 | — | — | |
| GPT-4V (0409)Model Version=04092024.12 | 67.5 | — | — | |
| BAGELParams=7B+7B, Types=AR+Diff, Architecture Category=Uni. MoE/MoT Arch2025.11 | 67.2 | — | — | |
| BAGELParams=7B2025.12 | 67.2 | — | — | |
| BAGELParams=7B+7B, Architecture category=Integrating Autoregressive and Diffusion within Transformers2026.03 | 67.2 | — | — | |
| BAGALType=Und., # Params=14B2025.03 | 67.2 | — | — | |
| BAGELParams=7B + 7B*2025.06 | 67.2 | — | — | |
| Bagel+LatentUMM2026.05 | 67.2 | — | — | |
| BAGELBaseline=BAGEL, Params=7B active2026.06 | 67.2 | — | — | |
| UniWorld-V1Params=7B+12B, Types=AR+Diff, Architecture Category=Uni. Frozen MLLM2025.11 | 67.1 | — | — | |
| RegularBase Model=Qwen2.5-VL2026.02 | 67.1 | — | — | |
| UniWorld-V1Params=7B2025.12 | 67.1 | — | — | |
| UniWorld-V1Params=7B + 12B, Architecture category=Multi-modal Understanding Models as Enhanced Conditional Encoders2026.03 | 67.1 | — | — | |
| UniWorld-V1Params=7B + 12B*2025.06 | 67.1 | — | — | |
| Qwen2.5-VLLLM=Qwen2.5-7B, Token Type=2D-Continuous dynamic2026.05 | 67.1 | — | — | |
| OPERABase Model=Qwen2.5-VL2026.02 | 66.9 | — | — | |
| MIRROR (ours)Param Size=7B2026.02 | 66.7 | — | — | |
| Qwen-VL-Maxopen-source=false2024.04 | 66.6 | — | — | |
| Qwen2.5-VLParams=7B, Types=AR, Architecture Category=Und. Only2025.11 | 66.6 | — | — | |
| MetaQuery-XLParams=7B+1.6B, Types=AR+Diff, Architecture Category=Uni. Frozen MLLM2025.11 | 66.6 | — | — | |
| Blip3-oParams=7B+1.4B, Types=AR+Diff, Architecture Category=Uni. Frozen MLLM2025.11 | 66.6 | — | — | |
| BLIP3-oParams=7B2025.12 | 66.6 | — | — | |
| MetaQueryParams=7B + 1.6B, Architecture category=Multi-modal Understanding Models as Enhanced Conditional Encoders2026.03 | 66.6 | — | — | |
| Blip3-oParams=7B + 1.4B, Architecture category=Multi-modal Understanding Models as Enhanced Conditional Encoders2026.03 | 66.6 | — | — | |
| MetaQuery-XLParams=7B + 1.6B*2025.06 | 66.6 | — | — | |
| BLIP3-o 8BParams=7B + 1.4B*2025.06 | 66.6 | — | — | |
| BLIP3-oLLM=Qwen2.5VL-7B-Instruct, Token Type=2D-Continuous dynamic2026.05 | 66.6 | — | — | |
| BLIP3o-8BBaseline=BLIP3o-8B, Params=8B2026.06 | 66.6 | — | — | |
| Bagel+RecA2026.05 | 66.1 | — | — | |
| Claude3.5-Sonnet2024.12 | 66 | — | — | |
| BagelParams=14B2026.05 | 65.9 | — | — | |
| Bagel+SFT2026.05 | 65.8 | — | — | |
| DoLaBase Model=Qwen2.5-VL2026.02 | 65.6 | — | — | |
| VLsIParameters=2B2024.12 | 64.8 | — | — | |
| VLSI-2BParameters=2B2024.12 | 64.8 | — | — | |
| VLSI-2BSize=2B2024.12 | 64.8 | — | — | |
| VCDBase Model=Qwen2.5-VL2026.02 | 64.7 | — | — | |
| Bagel+UniCot2026.05 | 64.5 | — | — | |
| InternVL2-76BSize=76B2024.12 | 64.4 | — | — | |
| Gemini-Pro2024.01 | 64.3 | — | — | |
| Gemini Pro 1.0open-source=false2024.04 | 64.3 | — | — | |
| GPT-4VLLM=Unk, Encoder=Unk, Resolution=Unk, Zero-shot=true2024.04 | 64.3 | — | — | |
| Gemini ProLLM=Unk, Encoder=Unk, Resolution=Unk, Zero-shot=true2024.04 | 64.3 | — | — | |
| InternVL3Param Size=8B2026.02 | 64.27 | — | — | |
| LLaMA-3.2-Vision2024.12 | 64.1 | — | — | |
| Gemini-1.5-Pro2024.12 | 64 | — | — | |
| Step-1V#param=100B, open-source=false2024.04 | 63.3 | — | — | |
| MUSE-3B# LLM Params=2B2026.05 | 62.9 | — | — | |
| InternVL 1.5#param=26B, open-source=false2024.04 | 62.8 | — | — | |
| InternVL2.5Params=8B, Types=AR, Architecture Category=Und. Only2025.11 | 62.8 | — | — | |
| FLARE-X 8B# Vis tok.=14002025.04 | 62.8 | — | — | |
| InternVL2.5LLM=InternLM2.5-7B, Token Type=2D-Continuous dynamic2026.05 | 62.8 | — | — | |
| Omnigen2Params=7B2026.05 | 62.7 | — | — | |
| InternVL3# LLM Params=1.8B2026.05 | 62.2 | — | — | |
| UniLIP-3B# LLM Params=2B2026.05 | 62.2 | — | — | |
| Qwen2.5 VLParams=7B, Architecture category=Unifying Understanding and Generation via Next-token Prediction, VLMEvalKit=true2026.03 | 62.1 | — | — | |
| FLARE-X 3B# Vis tok.=14002025.04 | 61.9 | — | — | |
| Qwen-VL-MAXLLM Size=Unk2024.03 | 61.8 | — | — | |
| Qwen-VL-MaxLLM=Unk, Encoder=Unk, Resolution=Unk, Zero-shot=true2024.04 | 61.8 | — | — | |
| Qwen2.5-VLParams=3B, Types=AR, Architecture Category=Und. Only2025.11 | 61.8 | — | — | |
| OmniGen2Params=3B+4B, Types=AR+Diff, Architecture Category=Uni. Frozen MLLM2025.11 | 61.8 | — | — | |
| OmniGen2Params=3B2025.12 | 61.8 | — | — | |
| OmniGen2Params=3B + 4B, Architecture category=Multi-modal Understanding Models as Enhanced Conditional Encoders2026.03 | 61.8 | — | — | |
| OmniGen2Params=3B + 4B*2025.06 | 61.8 | — | — | |
| Qwen2.5-VL# LLM Params=3B2026.05 | 61.8 | — | — | |
| Qwen2.5VL 3B# Vis tok.=14002025.04 | 61.4 | — | — | |
| Molmo-72BSize=72B2024.12 | 61.1 | — | — | |
| Qwen-VL-Plusopen-source=false2024.04 | 61.1 | — | — | |
| InternVL2.5# LLM Params=1.8B2026.05 | 60.8 | — | — | |
| Bagel+UniGame2026.05 | 60.7 | — | — | |
| LLaVA-OV-72BSize=72B2024.12 | 60.6 | — | — | |
| BLIP3o-4BType=Und. and Gen. > 2B, # Total Params=7.1B2026.02 | 60.1 | — | — | |
| BLIP3-o# LLM Params=4B2026.05 | 60.1 | — | — | |
| MUSE-1B# LLM Params=1B2026.05 | 60.1 | — | — | |
| MIRROR (w/o tool)Param Size=7B2026.02 | 59.91 | — | — | |
| FLARE-L 8B# Vis tok.=6302025.04 | 59.6 | — | — | |
| InternVL3# LLM Params=1B2026.05 | 59.5 | — | — | |
| UniLIP-1B# LLM Params=1B2026.05 | 59.4 | — | — | |
| Mini-Gemini#param=35B, open-source=false2024.04 | 59.3 | — | — |