List your company
ResearchFormerly WizWand

Multimodal Model Evaluation on MMBench Chinese

82.6AccuracySAIL-VL2
55.24862.34969.4576.551Nov 12, 2023Apr 1, 2024Aug 21, 2024Jan 10, 2025May 31, 2025Oct 20, 2025Mar 11, 2026
Updated 7mo ago

Evaluation results

MethodLinks
SAIL-VL2Number of Parameters=2.7B2025.1282.6
CoLLaVONumber of Parameters=7B, Setting=Zero-shot2024.0282.1
HyperVLNumber of Parameters=1.8B2025.1280.5
SAIL-VL1.5Number of Parameters=2.5B2025.1279.3
HyperVL ViTLNumber of Parameters=2.0B2025.1279.3
Ovis2Number of Parameters=2.5B2025.1278.9
AndesVLNumber of Parameters=2.4B2025.1278.6
InternVL3Number of Parameters=2.1B2025.1278.3
Qwen3-VLNumber of Parameters=2.1B2025.1278
InternVL3.5Number of Parameters=2.3B2025.1277.7
MoAIZero-shot=true, Number of Parameters=7B2024.0376.5
Qwen2-VLNumber of Parameters=2.2B2025.1274.2
LLaVA-XTunerNumber of Parameters=20B, Setting=Zero-shot2024.0273.7
LLaVA-XTunerZero-shot=true, Number of Parameters=20B2024.0373.7
Intern-XCNumber of Parameters=7B, Setting=Zero-shot2024.0272.4
Intern-XCZero-shot=true, Number of Parameters=7B2024.0372.4
VisionLLM v2-chatBackbone=Vicuna-7B, #Data=22M2024.0667.6
Dragonfly (Ours)Backbone=Llama3-8B, #Data=2.9M2024.0666.1
Merlin2023.1165.5
VILA-13B + ShareGPT4VLLM=Llama-2-13B, Resolution=336, Pre-training samples=50M, Instruction tuning samples=1M2023.1265.4
FMVR-LLaVA#Vision Tokens=576, Backbone=LLaVA-1.5-13B2026.0365.1
LLaVA-UHD#Data=1.2M, MaxRes.=672x1008, AR.=Any, TFLOPS=14.62024.0364.8
VILA-13BLLM=Llama-2-13B, Resolution=336, Pre-training samples=50M, Instruction tuning samples=1M2023.1264.3
ASMv2Model Scale=13B2024.0264.3
FMVR-LLaVA#Vision Tokens=144, Backbone=LLaVA-1.5-13B2026.0363.8
LLaVA-1.5#Data=1.2M, MaxRes.=336x336, AR.=Fix, TFLOPS=15.52024.0363.6
LLaVA-1.5LLM=Vicuna-1.5-13B, Resolution=336, Pre-training samples=0.6M, Instruction tuning samples=0.7M2023.1263.6
LLaVA1.5Number of Parameters=13B, Setting=Zero-shot2024.0263.6
LLaVA-v1.5-13BLLM=Vicuna-13B, Resolution=3362024.0563.6
LLaVA-1.5LLM=Vicuna-13B, Data=558K+665K2024.0163.6
CaMML-13BLLM=Vicuna-13B, Data=558K++665K2024.0163.6
LLaVA1.5Zero-shot=true, Number of Parameters=13B2024.0363.6
LLaVA-1.5Model Scale=13B2024.0263.6
LLaVA-1.5-13B#Vision Tokens=576, Backbone=LLaVA-1.5-13B2026.0363.5
LLaVA-1.5-13BResolution=336*3362024.1163.3
VILAModel Scale=13B2024.0262.8
PyramidDrop#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0362.8
SparseVLM#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0362.6
VisionZip#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0362.5
FMVR-LLaVA#Vision Tokens=36, Backbone=LLaVA-1.5-13B2026.0362.5
FastV#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0362.3
ShareGPT4VNumber of Parameters=7B, Setting=Zero-shot2024.0262.2
ShareGPT4VZero-shot=true, Number of Parameters=7B2024.0362.2
ShareGPT4V-7BBackbone=Vicuna-7B, #Data=1.8M2024.0662.2
DART#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0362.2
PPLLAVAResolution=336*3362024.1162
Qwen-VL-Chat2023.1161.8
VILA-7BLLM=Llama-2-7B, Resolution=336, Pre-training samples=50M, Instruction tuning samples=1M2023.1261.7
VILABackbone=Llama2-7B, #Data=61M2024.0661.7
CDPruner#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0361.5
VisPruner#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0361.3
Prumerge+#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0361.2
MQT-LLaVA#Vision Tokens=144, Backbone=LLaVA-1.5-13B2026.0361
DivPrune#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0360.7
HyperLLaVALLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0360.6
CaMML-7BLLM=Vicuna-7B, Data=558K++665K2024.0160.6
LLaVA-NeXTBackbone=Vicuna-7B, #Data=1.2M2024.0660.6
LLaVA-Next-7BResolution=672*6722024.1160.6
FMVR-LLaVA#Vision Tokens=9, Backbone=LLaVA-1.5-13B2026.0360.4
Shikra2023.1160.2
M3#Vision Tokens=144, Backbone=LLaVA-1.5-13B2026.0359.8
HyperLLaVA w/o EvLLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0359.7
LLaVA-1.52023.1159.5
InfMLLM-7B-ChatLLM=Vicuna-7B2023.1159.1
HyperLLaVA w/o ELLLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0358.6
TRIM#Vision Tokens=128, Backbone=LLaVA-1.5-13B2026.0358.4
LLaVA-1.5-7BLLM=Vicuna-7B2023.1158.3
LLaVA-1.5LLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0358.3
LLaVA-1.5LLM=Vicuna-1.5-7B, Resolution=336, Pre-training samples=0.6M, Instruction tuning samples=0.7M2023.1258.3
LLaVA1.5Number of Parameters=7B, Setting=Zero-shot2024.0258.3
LLaVA-1.5LLM=Vicuna-7B, Data=558K+665K2024.0158.3
LLaVA1.5Zero-shot=true, Number of Parameters=7B2024.0358.3
LLaVA-1.5Backbone=Vicuna-7B, #Data=1.2M2024.0658.3
FSRRetained Tokens=960, Reduction Ratio=↓ 66.7%2026.0258.3
VisPrunerRetained Tokens=960, Reduction Ratio=↓ 66.7%2026.0258.2
SPHINX-2k#Data=1.0B, MaxRes.=762x762, AR.=Fix, TFLOPS=69.42024.0357.9
SPHINX-2kBackbone=Llama2-7B, #Data=1B2024.0657.9
FSRRetained Tokens=640, Reduction Ratio=↓ 77.8%2026.0257.9
InternVL-7BBackbone=Vicuna-7B, #Data=>28.7B2024.0657.6
CDPrunerRetained Tokens=960, Reduction Ratio=↓ 66.7%2026.0257.6
CDPrunerRetained Tokens=640, Reduction Ratio=↓ 77.8%2026.0257.6
LLaVA-NeXT-7BRetained Tokens=2880, Reduction Ratio=100%2026.0257.4
DivPruneRetained Tokens=640, Reduction Ratio=↓ 77.8%2026.0257.3
VisPrunerRetained Tokens=640, Reduction Ratio=↓ 77.8%2026.0257.3
LLaVA-NeXT-7B#Vision Tokens=28802026.0357.3
FMVR-LLaVA#Vision Tokens=7202026.0357.1
HoloVRetained Tokens=320, Reduction Ratio=↓ 88.9%2026.0257
FMVR-LLaVA#Vision Tokens=28802026.0356.9
Qwen-VL-ChatLLM=Qwen-7B2023.1156.7
Qwen-VL-ChatLLM=Qwen-7B, Res.=448, PT=1.4B, IT=50M2024.0356.7
Qwen-VL-ChatLLM=Qwen-7B, Resolution=448, Pre-training samples=1.4B, Instruction tuning samples=50M2023.1256.7
Qwen-VL-ChatNumber of Parameters=7B, Setting=Zero-shot2024.0256.7
Qwen-VL-ChatLLM=Qwen-7B, Data=1.4B+50M2024.0156.7
Qwen-VL-ChatZero-shot=true, Number of Parameters=7B2024.0356.7
Qwen-VL-Chat2024.0256.7
Qwen-VL-ChatBackbone=Qwen-7B, #Data=1.4B2024.0656.7
LLaVA-Next-VideoResolution=336*3362024.1156.7
HoloVRetained Tokens=640, Reduction Ratio=↓ 77.8%2026.0256.7
SparseVLM#Vision Tokens=3202026.0356.7
FMVR-LLaVA#Vision Tokens=1802026.0356.3