List your company
ResearchFormerly WizWand

Multi-modal Understanding on LLaVA-Bench Wild

91.2LLaVA^W ScoreGPT4V
35.97650.31364.6578.987Nov 27, 2023Apr 19, 2024Sep 10, 2024Feb 1, 2025Jun 25, 2025Nov 16, 2025Apr 10, 2026
Updated 6mo ago

Evaluation results

MethodLinks
GPT4VPrompting Strategy=Compositional Chain-of-Thought (CCoT)2023.1191.2——
GPT4VPrompting Strategy=Zero-Shot CoT2023.1188.8——
Qwen-VL-Max + AutoVBase Model=Qwen-VL-Max, AutoV Integration=true, Retrieval Strategy Source=AutoV pre-trained on LLaVA-OneVision2025.0688.4——
GPT4VPrompting Strategy=Base2023.1188.2——
VILA2-8BLLM Parameters=8B, Vision Tower Parameters=400M, Tokens per image=196, Pre-training data size=51M2024.0786.6——
Qwen-VL-MaxBase Model=Qwen-VL-Max, AutoV Integration=false2025.0685.8——
ASPOBase Model=LLaVA-v1.5, Model Scale=13B, Optimization Protocol=ASPO2025.058268.8—
MM1-7B-ChatLLM Parameters=7B, Vision Tower Parameters=300M, Tokens per image=720, Pre-training data size=>2B2024.0781.5——
Gemini-1.5-Pro + AutoVBase Model=Gemini-1.5-Pro, AutoV Integration=true, Retrieval Strategy Source=AutoV pre-trained on LLaVA-OneVision2025.0681.1——
Gemini-1.5-ProBase Model=Gemini-1.5-Pro, AutoV Integration=false2025.0677.9——
Honeybeeprojector=C-Abstractor2023.1277.5——
DOP-OBCBase Model=LLaVA-1.52026.0477.3——
DOP-OBCBase Model=Video-LLaVA2026.0476.4——
DOP-OBCBase Model=Chat-UniVi2026.0475.9——
HoneybeeLLM=Vicuna-13B, Projector=C-Abstractor, Vision Encoder=CLIP ViT-L/14, Res.=336, Visual tokens (M)=2562023.1275.7——
ASPOBase Model=LLaVA-v1.5, Model Scale=7B, Optimization Protocol=ASPO2025.0575.765.16—
AlignGPTLLM=Vicuna-13B, Resolution=3362024.0575.2——
CAMDBase Model=LLaVA-1.52026.0375——
CAMDBase Model=Video-LLaVA2026.0375——
LLaVA-1.5-13BPrompting Strategy=Compositional Chain-of-Thought (CCoT)2023.1174.9——
DPOBase Model=LLaVA-v1.5, Model Scale=13B, Optimization Protocol=DPO2025.0574.767.03—
FarSightBase Model=LLaVA-1.52026.0374.7——
FarSightBase Model=LLaVA-1.52026.0474.7——
FarSightBase Model=Video-LLaVA2026.0374.5——
CAMDBase Model=Chat-UniVi2026.0373.6——
LLaVA-1.5-13BPrompting Strategy=Base2023.1173.5——
Video-LLaVADecoding Strategy=Greedy2026.0373.1——
BaseBase Model=Video-LLaVA2026.0473.1——
HoneybeeLLM=Vicuna-13B, Projector=D-Abstractor, Vision Encoder=CLIP ViT-L/14, Res.=336, Visual tokens (M)=2562023.1272.9——
FarSightBase Model=Chat-UniVi2026.0372.6——
LLaVA-1.5Decoding Strategy=Greedy2026.0372.5——
BaseBase Model=LLaVA-1.52026.0472.5——
OPERABase Model=LLaVA-1.52026.0372——
OPERABase Model=LLaVA-1.52026.0472——
VisionZipBase Model=LLaVA 1.5 7B, Number of Visual Tokens=1922024.1271.3—106.7
CGDBase Model=LLaVA-1.52026.0371.3——
SphinxPrompting Strategy=Compositional Chain-of-Thought (CCoT)2023.1171——
GPT-4o + AutoVBase Model=GPT-4o, AutoV Integration=true, Retrieval Strategy Source=AutoV pre-trained on LLaVA-OneVision2025.0670.9——
VCDBase Model=LLaVA-1.52026.0370.9——
VCDBase Model=LLaVA-1.52026.0470.9——
LLaVA-1.5LLM=Vicuna-13B, Projector=Linear, Vision Encoder=CLIP ViT-L/14, Res.=3362023.1270.7——
LLaVA-v1.52023.1270.7——
LLaVA-1.5-13BLLM=Vicuna-1.5-13B, Resolution=3362024.0670.7——
LLaVA-1.5LLM=Vicuna-13B, Resolution=3362024.0570.7——
LLaVA-1.5-13BBase Model=LLaVA-v1.5, Model Scale=13B, Optimization Protocol=Base2025.0570.765.93—
Chat-UniViDecoding Strategy=Greedy2026.0370.4——
BaseBase Model=Chat-UniVi2026.0470.4——
LLaVA-Instruct-13BLLM=Vicuna-1.5-13B, Resolution=3362024.0670.1——
SphinxPrompting Strategy=Base2023.1170——
SphinxPrompting Strategy=Zero-Shot CoT2023.1169.8——
ICDBase Model=LLaVA-1.52026.0369.7——
ICDBase Model=LLaVA-1.52026.0469.7——
LLaVA-1.5-13BPrompting Strategy=Zero-Shot CoT2023.1168.5——
AlignGPTLLM=Vicuna-7B, Resolution=3362024.0568.4——
GPT-4oBase Model=GPT-4o, AutoV Integration=false2025.0668——
ASPOBase Model=InstructBLIP, Model Scale=13B, Optimization Protocol=ASPO2025.0567.445.84—
HoneybeeLLM=Vicuna-7B, Projector=C-Abstractor, Vision Encoder=CLIP ViT-L/14, Res.=224, Visual tokens (M)=1442023.1267.1——
VanillaBase Model=LLaVA 1.5 7B, Number of Visual Tokens=5762024.1266.8—100
VisionZipBase Model=LLaVA 1.5 7B, Number of Visual Tokens=1282024.1266.7—99.9
HoneybeeLLM=Vicuna-7B, Projector=D-Abstractor, Vision Encoder=CLIP ViT-L/14, Res.=224, Visual tokens (M)=1442023.1266.3——
DPOBase Model=LLaVA-v1.5, Model Scale=7B, Optimization Protocol=DPO2025.0565.763.12—
LLaVA-Instruct-7BLLM=Vicuna-1.5-7B, Resolution=3362024.0665.1——
DOP-OBCBase Model=InstructBLIP2026.0464.3——
VisionZipBase Model=LLaVA 1.5 7B, Number of Visual Tokens=642024.1263.5—95.1
LLaVA-1.5LLM=Vicuna-7B, Projector=Linear, Vision Encoder=CLIP ViT-L/14, Res.=3362023.1263.4——
LLaVA-1.5-7BLLM=Vicuna-1.5-7B, Resolution=3362024.0663.4——
LLaVA-1.5LLM=Vicuna-7B, Resolution=3362024.0563.4——
LLaVA-1.5-7BBase Model=LLaVA-v1.5, Model Scale=7B, Optimization Protocol=Base2025.0563.462.59—
LLaVA-7BBase Model=LLaVA, Model Scale=7B, Optimization Protocol=Base2025.0563——
DPOBase Model=InstructBLIP, Model Scale=13B, Optimization Protocol=DPO2025.0562.744.5—
CAMDBase Model=InstructBLIP2026.0361.2——
FarSightBase Model=InstructBLIP2026.0361——
InstructBLIPLLM=Vicuna-7B, Projector=Q-former, Vision Encoder=EVA-CLIP ViT-G, Res.=2242023.1260.9——
InstructBLIP-7BLLM=Vicuna-7B, Resolution=2242024.0660.9——
InstructBLIPLLM=Vicuna-7B, Resolution=2242024.0560.9——
InstructBLIPLLM=Vicuna-13B, Projector=Q-former, Vision Encoder=EVA-CLIP ViT-G, Res.=2242023.1258.2——
InstructBLIPLLM=Vicuna-13B, Resolution=2242024.0558.2——
InstructBLIP-13BBase Model=InstructBLIP, Model Scale=13B, Optimization Protocol=Base2025.0558.243.86—
InstructBLIPDecoding Strategy=Greedy2026.0358.2——
BaseBase Model=InstructBLIP2026.0458.2——
InstructBLIP-13BPrompting Strategy=Compositional Chain-of-Thought (CCoT)2023.1147.9——
InstructBLIP-13BPrompting Strategy=Base2023.1147.2——
InstructBLIP-13BPrompting Strategy=Zero-Shot CoT2023.1145.4——
BLIP-2LLM=Vicuna-13B, Projector=Q-former, Vision Encoder=EVA-CLIP ViT-G, Res.=2242023.1238.1——
BLIP-2LLM=Vicuna-13B, Resolution=2242024.0538.1——
BLIP-2-7BBase Model=BLIP-2, Model Scale=7B, Optimization Protocol=Base2025.0538.1——