List your company
ResearchFormerly WizWand

Multimodal Reasoning on MMBench (dev)

87.6AccuracyGPT-4o
1.2823.6946.168.51Dec 28, 2023May 3, 2024Sep 7, 2024Jan 12, 2025May 19, 2025Sep 23, 2025Jan 28, 2026
Updated 8mo ago

Evaluation results

MethodLinks
GPT-4oindependent reproduction=true, IoT prompting=true2024.0587.6
GPT-4oindependent reproduction=true, IoT prompting=false2024.0586.2
Gemini-Pro-1.5independent reproduction=true, IoT prompting=true2024.0579.6
InternLM-XComposer2-VLParams=7B2024.0579.5
Qwen-VL-Max2024.0578.1
LLaVA-FA-7BLLM=InternLM-2-20B, #Sample=5M, #Param=≥7B2026.0174.5
Deepseek-VL-7BLLM=DLLM-7B, #Sample=2000M, #Param=≥7B2026.0174.1
Imp-3BLLM=Phi-2-2.7B, #Sample=1.6M, #Param=∼3B2026.0172.9
Gemini-Pro-1.5independent reproduction=true, IoT prompting=false2024.0571
LLaVA-FA-3BLLM=LLaMA-3-8B, #Sample=5M, #Param=∼3B2026.0170.5
LLaVA-NeXTLLM=Vicuna-1.5-13B, #Sample=1.3M, #Param=≥7B2026.0170.4
LLaVA-v1.5-13BParams=13.4B2024.0569.2
VILA-7BLLM=LLaMA-7B, #Sample=50M, #Param=≥7B2026.0168.9
Bunny-3BLLM=Phi-2-2.7B, #Sample=2.7M, #Param=∼3B2026.0168.6
MiniCPM-V-2LLM=MiniCPM-2.4B, #Sample=570M, #Param=∼3B2026.0168.5
LLaVA-FA-2BLLM=Qwen-2.5-7B, #Sample=5M, #Param=∼2B2026.0166.7
Qwen-VL-Plus2024.0566.2
MoE-LLaVA-3BLLM=Phi-2-2.7B, #Sample=2.2M, #Param=∼3B2026.0165.2
DeepSeek-VL-1.3BLLM=DLLM-1.3B, #Sample=2000M, #Param=∼2B2026.0164.6
LLaVA-1.5LLM=Vicuna-7B, Resolution (Res.)=336, Pre-training data size (PT)=558K, Instruction tuning data size (IT)=665K2023.1264.3
LLaVA-1.5-7BLLM=Vicuna-1.5-7B, #Sample=1.2M, #Param=≥7B2026.0164.3
MiniCPM-VLLM=MiniCPM-2.4B, #Sample=570M, #Param=∼3B2026.0164
Imp-2BLLM=Qwen-1.5-1.8B, #Sample=1.6M, #Param=∼2B2026.0163.8
CogVLMLLM=Vicuna-7B, #Sample=1500M, #Param=≥7B2026.0163.7
VILA-3BLLM=LLaMA-2.7B, #Sample=51M, #Param=∼3B2026.0163.4
MobileVLMv2LLM=MLLaMA-2.7B, #Sample=3.6M, #Param=∼3B2026.0163.2
Qwen-VL-ChatLLM=Qwen-7B, #Sample=1450M, #Param=≥7B2026.0160.6
Mini-Gemini-2BLLM=Gemma-2B, #Sample=2.7M, #Param=∼2B2026.0159.8
MoE-LLaVA-2BLLM=Qwen-1.5-1.8B, #Sample=2.2M, #Param=∼2B2026.0159.7
MobileVLM 3BLLM=MobileLLaMA 2.7B, Resolution (Res.)=336, Pre-training data size (PT)=558K, Instruction tuning data size (IT)=665K2023.1259.6
MobileVLMLLM=MLLaMA-2.7B, #Sample=1.3M, #Param=∼3B2026.0159.6
Bunny-2BLLM=Qwen-1.5-1.8B, #Sample=2.7M, #Param=∼2B2026.0159.1
ShikraLLM=Vicuna-13B, Resolution (Res.)=224, Pre-training data size (PT)=600K, Instruction tuning data size (IT)=5.5M2023.1258.8
LLaVA-FA-1BLLM=Qwen-2.5-3B, #Sample=5M, #Param=∼1B2026.0158.3
MobileVLM 3B w/ LORALLM=MobileLLaMA 2.7B, Resolution (Res.)=336, Pre-training data size (PT)=558K, Instruction tuning data size (IT)=665K2023.1257
SPHINX-TinyLLM=TLLaMA-1.1B, #Sample=15M, #Param=∼1B2026.0156.6
IDEFICS-80BLLM=LLaMA-65B, Resolution (Res.)=224, Pre-training data size (PT)=353M, Instruction tuning data size (IT)=1M2023.1254.5
MobileVLM 1.7BLLM=MobileLLaMA 1.4B, Resolution (Res.)=336, Pre-training data size (PT)=558K, Instruction tuning data size (IT)=665K2023.1253.2
MobileVLM 1.7B w/ LORALLM=MobileLLaMA 1.4B, Resolution (Res.)=336, Pre-training data size (PT)=558K, Instruction tuning data size (IT)=665K2023.1250.4
mPLUG-OwlLLM=LLaMA-7B, Resolution (Res.)=224, Pre-training data size (PT)=2.1M, Instruction tuning data size (IT)=102K2023.1249.4
IDEFICS-9BLLM=LLaMA-7B, Resolution (Res.)=224, Pre-training data size (PT)=353M, Instruction tuning data size (IT)=1M2023.1248.2
Qwen-VLLLM=Qwen-7B, Resolution (Res.)=448, Pre-training data size (PT)=1.4B, Instruction tuning data size (IT)=50M2023.1238.2
InstructBLIPLLM=Vicuna-7B, Resolution (Res.)=224, Pre-training data size (PT)=129M, Instruction tuning data size (IT)=1.2M2023.1236
MiniGPT-4Params=8B2024.0524.3
MiniGPT-4LLM=Vicuna-7B, Resolution (Res.)=224, Pre-training data size (PT)=5M, Instruction tuning data size (IT)=5K2023.1223
MiniGPT-v2LLM=LLaMA-7B, Resolution (Res.)=448, Pre-training data size (PT)=23M, Instruction tuning data size (IT)=1M2023.1212.2
OpenflamingoLLM=MPT-7B, Resolution (Res.)=336, Pre-training data size (PT)=180M, Instruction tuning data size (IT)=-2023.124.6