Multimodal Understanding on SEED
82.6AccuracyInternVL3-8B-Masters
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InternVL3-8B-MastersSize=8B2025.12 | 82.6 | — | |
| MastersBase Model=InternVL3-8B2025.12 | 82.6 | — | |
| SPHINX-TinyModel Scale=1B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 82.2 | — | |
| Qwen2.5-VL-7B-MastersSize=7B2025.12 | 81.8 | — | |
| MastersBase Model=Qwen2.5-VL-7B2025.12 | 81.8 | — | |
| Qwen3-VL-8B-MastersSize=8B2025.12 | 81.7 | — | |
| MastersBase Model=Qwen3-VL-8B2025.12 | 81.7 | — | |
| Qwen3-VL-4B-MastersSize=4B2025.12 | 81.5 | — | |
| InternVL3.5-8B-MastersSize=8B2025.12 | 81.4 | — | |
| MastersBase Model=InternVL3.5-8B2025.12 | 81.4 | — | |
| Qwen3-VL-32B2025.12 | 79.9 | — | |
| Qwen2.5-VL-72B2025.12 | 79.5 | — | |
| InternVL3.5-4B-MastersSize=4B2025.12 | 79.1 | — | |
| InternVL3.5-38B2025.12 | 79.1 | — | |
| InternVL3.5-2B-MastersSize=2B2025.12 | 79 | — | |
| Qwen2.5-VL-3B-MastersSize=3B2025.12 | 78.8 | — | |
| InternVL3-78B2025.12 | 78.7 | — | |
| Qwen3-VL-8BSize=8B2025.12 | 78.5 | — | |
| InternVL3-2B-MastersSize=2B2025.12 | 78.5 | — | |
| BAGEL#Params=7B+7B2026.03 | 78.5 | — | |
| Qwen3-VL-4BSize=4B2025.12 | 78.4 | — | |
| Qwen3-VL-2B-MastersSize=2B2025.12 | 78.3 | — | |
| GPT-4.12025.12 | 78 | — | |
| LLaVA-OneVision-72B2025.12 | 77.5 | — | |
| InternVL3.5-8BSize=8B2025.12 | 77.4 | — | |
| LLaVA-OneVision-1.5-8BSize=8B2025.12 | 77.3 | — | |
| Ovis2-8BSize=8B2025.12 | 77.2 | — | |
| GPT-4oZero-shot=true, Decoding Strategy=greedy2024.09 | 77.1 | — | |
| InternVL3-8BSize=8B2025.12 | 77.1 | — | |
| GPT-4o2025.12 | 77.1 | — | |
| Qwen2.5-VL-7BSize=7B2025.12 | 77 | — | |
| FastVtoken ratio=64%2024.12 | 76.81 | 99.84 | |
| InternVL2-26Btoken ratio=100%2024.12 | 76.78 | 100 | |
| SGPtoken ratio=64%2024.12 | 76.71 | 100.14 | |
| LLaVA-OneVision-7BSize=7B2025.12 | 76.7 | — | |
| LLaVA-OneVision-1.5-4BSize=4B2025.12 | 76.6 | — | |
| Vanilla2026.04 | 76.6 | — | |
| InternVL3.5 + REGATELLM=Qwen2-7B2025.07 | 76.6 | — | |
| Ovis2-4BSize=4B2025.12 | 76.2 | — | |
| Qwen3-VL-2BSize=2B2025.12 | 76 | — | |
| Gemini-1.5-Pro2025.12 | 76 | — | |
| LLaVA-NeXT-34BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 75.9 | — | |
| SGPtoken ratio=35%2024.12 | 75.72 | 98.36 | |
| ToMetoken ratio=64%2024.12 | 75.6 | 94.24 | |
| MiniCPM-o2.6-8BSize=8B2025.12 | 75.5 | — | |
| NVLM-72B2025.12 | 75.5 | — | |
| Ovis-U1#Params=2.4B+1.2B2026.03 | 75.5 | — | |
| Cambrian-34BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 75.3 | — | |
| InternVL3.5#Params=2B2026.03 | 75.3 | — | |
| InternVL3.5-4BSize=4B2025.12 | 75.2 | — | |
| InternVL3.5-2BSize=2B2025.12 | 75.2 | — | |
| InternVL-U#Params=2B+1.7B2026.03 | 75.2 | — | |
| MM1.5-30BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 75 | — | |
| InternVL3-2BSize=2B2025.12 | 75 | — | |
| Molmo-72B2025.12 | 74.6 | — | |
| Ovis2-2BSize=2B2025.12 | 74.4 | — | |
| Claude-3.7-Sonnet2025.12 | 74.3 | — | |
| MiniCPM-V2.6-8BSize=8B2025.12 | 74 | — | |
| Qwen2.5-VL-3BSize=3B2025.12 | 74 | — | |
| Aquila-VL-2BSize=2B2025.12 | 73.9 | — | |
| Eagle-X3PT=595 K, SFT=1.8 M2025.01 | 73.9 | — | |
| ToMetoken ratio=35%2024.12 | 73.84 | 85.2 | |
| BLIP3o-4BType=Und. and Gen. > 2B, # Total Params=7.1B2026.02 | 73.8 | — | |
| MetaQuery-L#Params=3B+1B2026.03 | 73.8 | — | |
| Qwen2.5-VL#Params=3B2026.03 | 73.7 | — | |
| Eagle-X2PT=595 K, SFT=1.8 M2025.01 | 73.5 | — | |
| MM1.5-7BModel Scale=7B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 73.4 | — | |
| MM1.5-3B-MoEModel Scale=3B, Architecture=MoE, Zero-shot=true, Decoding Strategy=greedy2024.09 | 73.3 | — | |
| Phi-4-Multimodal-5.6BSize=5.6B2025.12 | 73.2 | — | |
| ToFuViT retain tokens=50%, LLM retain tokens=50%2026.04 | 73.2 | — | |
| SplitQBackbone=Qwen2.5-VL-7B, Bits=W4A82026.05 | 73.2 | — | |
| FP16Backbone=Qwen2.5-VL-7B, Bits=W16A162026.05 | 73 | — | |
| AIMv2architecture=ViT-3B/14, #patches=576, resolution=336px2024.11 | 72.9 | — | |
| ILLUMELLM=Vicuna-7B, Category=Unify Understanding and Generation2024.12 | 72.9 | — | |
| ToMeViT retain tokens=50%, LLM retain tokens=50%2026.04 | 72.8 | — | |
| AIMv2architecture=ViT-1B/14, #patches=576, resolution=336px2024.11 | 72.7 | — | |
| SplitQBackbone=Qwen2.5-VL-7B, Bits=W4A42026.05 | 72.6 | — | |
| BunnyModel Scale=3B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 72.5 | — | |
| MM1.5-3BModel Scale=3B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 72.4 | — | |
| Qwen2-VL-2B#Params=2B, TFLOPs=0.42025.06 | 72.4 | — | |
| LaTtE-Flow#Params=2B, TFLOPs=0.42025.06 | 72.4 | — | |
| SGPtoken ratio=9%2024.12 | 72.23 | 89.58 | |
| BLIP-3Model Scale=3B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 72.2 | — | |
| LeoPT=595 K, SFT=1 M2025.01 | 72.2 | — | |
| MM1-30BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 72.1 | — | |
| AIMv2architecture=ViT-H/14, #patches=576, resolution=336px2024.11 | 72.1 | — | |
| Phi-3-Vision-4BModel Scale=3B, Zero-shot=true, Decoding Strategy=greedy2024.09 | 71.8 | — | |
| AIMv2architecture=ViT-L/14, #patches=576, resolution=336px2024.11 | 71.8 | — | |
| MetaMorph#Params=8B, TFLOPs=1.12025.06 | 71.8 | — | |
| ToFuViT retain tokens=35%, LLM retain tokens=35%2026.04 | 71.7 | — | |
| FastVtoken ratio=35%2024.12 | 71.64 | 88.28 | |
| InternVL2-2Btoken ratio=100%2024.12 | 71.62 | 87.25 | |
| GPT-4VZero-shot=true, Decoding Strategy=greedy2024.09 | 71.6 | — | |
| ToMeViT retain tokens=35%, LLM retain tokens=35%2026.04 | 71.5 | — | |
| MM1.5-1B-MoEModel Scale=1B, Architecture=MoE, Zero-shot=true, Decoding Strategy=greedy2024.09 | 71.4 | — | |
| SplitQBackbone=Qwen2.5-VL-7B, Bits=W3A32026.05 | 71.1 | — | |
| InternVL2-2BModel Scale=1B, Zero-shot=true, Decoding Strategy=beam search2024.09 | 70.9 | — | |
| DASHViT retain tokens=50%, LLM retain tokens=50%2026.04 | 70.9 | — | |
| JanusFlowType=Und. and Gen. ≤ 2B, # Total Params=2.1B2026.02 | 70.5 | — | |
| JanusFlow#Params=1.3B2026.03 | 70.5 | — |