List your company
ResearchFormerly WizWand

Visual Question Answering on VQAv2

97.4AccuracyOracle
77.74482.84787.9593.053Dec 12, 2023May 15, 2024Oct 17, 2024Mar 21, 2025Aug 23, 2025Jan 25, 2026Jun 30, 2026
Updated 4mo ago

Evaluation results

MethodLinks
OracleStrategy=Oracle2026.0597.4
SEA-PRIMELLM=Llama3-8B, Res.=3842024.0883.1
MM1Vision Encoder=CLIP-L, Resolution=1792, LLM=7B, Data=1B2025.0182.8
PIIP-LLaVAVision Encoder=ConvNeXt-L, CLIP-L, Resolution=1024/336, LLM=Vicuna-13B, Data=1.2M2025.0182.5
LLAVA-HRVision Encoder=ConvNeXt-L, CLIP-L, Resolution=1024/448, LLM=Vicuna-13B, Data=1.2M2025.0182.3
PIIP-LLaVAVision Encoder=ConvNeXt-L, CLIP-L, Resolution=1024/336, LLM=Vicuna-7B, Data=2.7M2025.0182.3
CogVLM# Param=17B2024.0882.3
MM1-7B-ChatBackbone=MM1-7B, #Data=>2B2024.0682.3
mPLUG-Owl3# Param=8B2024.0882.1
MG-LLAVALLM=Yi1.5-34B, Parameters=34.4B, Resolution=336(768)2024.0682
LLaVA-HRVision Encoder=ConvNeXt-L, CLIP-L, Resolution=1024/448, LLM=Vicuna-7B, Data=1.2M2025.0181.9
SEA-PRIMELLM=Vicuna-13B, Res.=3842024.0881.9
EVLM-Chat# Param=32B2024.0881.9
FP16Backbone=Qwen3-VL-8B2026.0681.84
PIIP-LLaVAVision Encoder=ConvNeXt-B, CLIP-L, Resolution=1024/336, LLM=Vicuna-13B, Data=1.2M2025.0181.8
LLaVA-NeXTVision Encoder=CLIP-L, Resolution=1344, LLM=Vicuna-7B, Data=1.6M2025.0181.8
LLAVA-Next# Param=8B2024.0881.8
LLaVA-NeXTBackbone=Vicuna-7B, #Data=1.2M2024.0681.8
LLaVA-UHDLLM=Vicuna-13B, Parameters=13.4B, Resolution=672x10082024.0681.7
PIIP-LLaVAVision Encoder=ConvNeXt-L, CLIP-L, Resolution=1024/336, LLM=Vicuna-7B, Data=1.2M2025.0181.5
SEA-PRIMELLM=Vicuna-7B, Res.=3842024.0881.4
VisionLLM v2-chatBackbone=Vicuna-7B, #Data=22M2024.0681.4
Flamingo# Param=80B2024.0881.3
Mipha-3BLM=Phi-2-2.7B, Res.=3842024.1281.3
MG-LLAVALLM=Vicuna-13B, Parameters=13.6B, Resolution=336(768)2024.0681.2
PIIP-LLaVAVision Encoder=ConvNeXt-B, CLIP-L, Resolution=1024/336, LLM=Vicuna-7B, Data=1.2M2025.0181.1
SharQBackbone=Qwen3-VL-8B2026.0681.1
SEA-PRIMELLM=Gemma-2B, Res.=3842024.0881
Dragonfly (Ours)Backbone=Llama3-8B, #Data=2.9M2024.0681
S2-WrapperLLM=Vicuna-13B, Res.=10082024.0880.9
VILA-1.5# Param=8B2024.0880.9
VILA-13BLLM=Llama-2-13B, Resolution=336, Pre-training samples=50M, Instruction tuning samples=1M2023.1280.8
Idefics2# Param=8B2024.0880.8
DualPDBackbone=Qwen-2.5-VL-7B, Decoding Strategy=DualPD2026.0180.78
SEA-PRIMELLM=Phi3-3.8B, Res.=3842024.0880.7
MG-LLAVALLM=LLaMA3-8B, Parameters=8.2B, Resolution=336(768)2024.0680.7
SPHINX-2kBackbone=Llama2-7B, #Data=1B2024.0680.7
ShareGPT4VBackbone=Vicuna-7B, Resolution=336, Visual Input Type=Continuous2024.1280.6
ShareGPT4VLLM=Vicuna-7B, Res.=3362024.0880.6
VILA-13B + ShareGPT4VLLM=Llama-2-13B, Resolution=336, Pre-training samples=50M, Instruction tuning samples=1M2023.1280.6
ShareGPT4V-7BBackbone=Vicuna-7B, #Data=1.8M2024.0680.6
Cloud-OnlyStrategy=Cloud-Only2026.0580.6
OlympusLM=Phi-2-2.7B, Res.=3842024.1280.5
NVFP4Backbone=Qwen3-VL-8B2026.0680.47
SlimeVision Encoder=CLIP-L, Resolution=2016, LLM=Vicuna-7B, Data=2M2025.0180.3
MonkeyBackbone=Qwen-7B, #Data=1B2024.0680.3
MG-LLaVAVision Encoder=ConvNeXt-L, CLIP-L, RAMPlus, OWL-ViTv2-L, Resolution=768/336, LLM=Vicuna-7B, Data=2.5M2025.0180.2
SPHINX-1kLLM=Vicuna-7B, Parameters=10B, Resolution=4482024.0680.2
MG-LLAVALLM=Vicuna-7B, Parameters=7.4B, Resolution=336(768)2024.0680.2
HiMAPModel=LLaVA-13B, TFLOPS=1.36, FLOPs Ratio=23%2025.0380.2
LLaVA-1.5-13B + MMFuserLLM=Vicuna-13B, Training annotations observed=true2024.1080.1
LLaVA-1.5Vision Encoder=CLIP-L, Resolution=336, LLM=Vicuna-13B, Data=1.2M2025.0180
LLaVA-1.5Backbone=Vicuna-13B, Resolution=336, Visual Input Type=Continuous2024.1280
Show-oBackbone=Phi-1.5-1.3B, Resolution=256, Visual Input Type=Discrete2024.1280
LLaMA-VIDLLM=Vicuna-13B, Res.=3362024.0880
LLaVA-1.5*LLM=Vicuna-13B, Res.=3362024.0880
AlignGPTLLM=Vicuna-13B, Res.=3362024.0880
LLaVA-1.5LLM=Vicuna-1.5-13B, Resolution=336, Pre-training samples=0.6M, Instruction tuning samples=0.7M2023.1280
LLaVA-1.5LLM=Vicuna-13B, Parameters=13.4B, Resolution=3362024.0680
LLaVA-1.5-13BLLM=Vicuna-13B, Training annotations observed=true2024.1080
Full-FinetuneTarget Model=LLaVA-1.5-13B, Data Budget=100%2026.0580
LLaVA-1.5-13BBackbone=LLaVA-1.5-13B, Retained Tokens=576 (Upper Bound)2026.0680
TinyLLaVALLM=Phi-2.7B, Res.=3842024.0879.9
VILA-7BLLM=Llama-2-7B, Resolution=336, Pre-training samples=50M, Instruction tuning samples=1M2023.1279.9
VILABackbone=Llama2-7B, #Data=61M2024.0679.9
MoE-LLaVA-3.6BLM=Phi-2-2.7B, Res.=3842024.1279.9
TinyLLaVALM=Phi-2-2.7B, Res.=3842024.1279.9
FastVModel=LLaVA-13B, TFLOPS=3.09, FLOPs Ratio=53%2025.0379.9
Visual PromptLLM=Vicuna-7B, Res.=3362024.0879.8
Bunny-3BLM=Phi-2-2.7B, Res.=3842024.1279.8
BaselineModel=LLaVA-13B, TFLOPS=5.81, FLOPs Ratio=100%2025.0379.8
S2-Wrapper*LLM=Vicuna-7B, Res.=10082024.0879.7
HiMAPModel=InternVL-7B, TFLOPS=0.56, FLOPs Ratio=20%2025.0379.6
Qwen-VLLLM=Qwen-7B, Res.=4482024.0879.5
Imp-v1LM=Phi-2-2.7B, Res.=3842024.1279.5
mPLUG-Owl2Vision Encoder=CLIP-L, Resolution=448, LLM=Llama-2-7B, Data=400M2025.0179.4
VILA-UBackbone=LLaMA-2-7B, Resolution=384, Visual Input Type=Discrete2024.1279.4
mPLUG-Owl2# Param=8B2024.0879.4
mPLUG-Owl2Backbone=Llama2-7B, #Data=401M2024.0679.4
mPLUG-Owl2LM=LLaMA-7B, Res.=4482024.1279.4
BaselineModel=InternVL-7B, TFLOPS=2.71, FLOPs Ratio=100%2025.0379.4
LLaMA-VIDLLM=Vicuna-7B, Res.=3362024.0879.3
InternVL-7BBackbone=Vicuna-7B, #Data=>28.7B2024.0679.3
Qwen-2.5-VL-7BBackbone=Qwen-2.5-VL-7B, Decoding Strategy=Standard2026.0179.3
AlignGPTLLM=Vicuna-7B, Res.=3362024.0879.1
LLaVA-1.5-7B + MMFuserLLM=Vicuna-7B, Training annotations observed=true2024.1079.1
INAR-VLStrategy=INAR-VL2026.0579.1
FastVModel=InternVL-7B, TFLOPS=1.39, FLOPs Ratio=52%2025.0379
MAGICTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0578.9
DoLABackbone=Qwen-2.5-VL-7B, Decoding Strategy=DoLA2026.0178.85
LLaVA-1.5*LLM=Vicuna-7B, Res.=3362024.0878.8
Qwen-VLLLM=Qwen-7B, Resolution=448, Pre-training samples=1.4B, Instruction tuning samples=50M2023.1278.8
Qwen-VLLLM=Qwen-7B, Parameters=9.6B, Resolution=4482024.0678.8
Qwen-VLLLM=Qwen-7B, Training annotations observed=true2024.1078.8
Qwen-VLLLM=Qwen-7B2024.1178.8
HiMAPModel=QwenVL-7B, TFLOPS=0.89, FLOPs Ratio=25%2025.0378.8
Text-OnlyStrategy=Text-Only2026.0578.8
HiMAPModel=LLaVA-7B, TFLOPS=0.73, FLOPs Ratio=24%2025.0378.6
DoLABackbone=Qwen-2-VL-7B, Decoding Strategy=DoLA2026.0178.57
LLaVA-1.5Vision Encoder=CLIP-L, Resolution=336, LLM=Vicuna-7B, Data=1.2M2025.0178.5