List your company
ResearchFormerly WizWand

Visual Question Answering on VQA v2 (Performance Score)

83.93AccuracyVisionZero-CLEVR
76.098878.131980.16582.1981Jun 2, 2026Jun 8, 2026Jun 14, 2026Jun 20, 2026Jun 26, 2026Jul 2, 2026Jul 8, 2026
Updated 3mo ago

Evaluation results

MethodLinks
VisionZero-CLEVRModel Scale=32B2026.0683.93
VisionZero-ChartModel Scale=32B2026.0683.9
VISEModel Scale=32B2026.0683.85
iReasonerModel Scale=32B2026.0683.84
EvoLMMModel Scale=32B2026.0683.8
VisionZero-RWModel Scale=32B2026.0683.61
BaseModel Scale=32B2026.0683.56
VisPlayModel Scale=32B2026.0683.52
BLIP-3oUnified=✓, #Params=8B, Visual Representation Type=Continuous2026.0683.1
Qwen2.5-VL-7B (Base)Base Model=Qwen2.5-VL-7B, Alignment Strategy=Base2026.0682.92
LLaVA-NeXT-13B + MIRRORModel Scale=13B-class, LLM=Vicuna-13B2026.0682.9
LLaVA-NeXT-13BModel Scale=13B-class, LLM=Vicuna-13B2026.0682.8
VisionZero-CLEVRModel Scale=8B2026.0682.72
VisionZero-ChartModel Scale=8B2026.0682.69
VISEModel Scale=8B2026.0682.65
iReasonerModel Scale=8B2026.0682.46
EvoLMMModel Scale=8B2026.0682.38
Hyper-ICLBackbone=LLaVA-OneVision (Qwen2-7B)2026.0682.24
LLaVA-NeXT-7BModel Scale=7B-class, LLM=Mistral-7B2026.0682.2
LLaVA-NeXT-7B + MIRRORModel Scale=7B-class, LLM=Mistral-7B2026.0682.1
VisionZero-RWModel Scale=8B2026.0681.92
BaseModel Scale=8B2026.0681.81
VisPlayModel Scale=8B2026.0681.78
LLaVA-1.5-13B + MIRRORModel Scale=13B-class, LLM=Vicuna-13B2026.0681.7
GRPOBase Model=Qwen2.5-VL-7B, Alignment Strategy=GRPO2026.0681.65
LLaVA-1.5-13BModel Scale=13B-class, LLM=Vicuna-13B2026.0681.3
MimICBackbone=LLaVA-OneVision (Qwen2-7B)2026.0681.22
Qwen3-VL-4B (Base)Base Model=Qwen3-VL-4B, Alignment Strategy=Base2026.0681.19
VISEModel Scale=4B2026.0681.16
iReasonerModel Scale=4B2026.0681.1
LLaVA-1.5-7B + MIRRORModel Scale=7B-class, LLM=Vicuna-7B2026.0681.1
VisionZero-ChartModel Scale=4B2026.0681.09
VisionZero-CLEVRModel Scale=4B2026.0681.07
EvoLMMModel Scale=4B2026.0680.97
OPDBase Model=Qwen3-VL-4B, Alignment Strategy=OPD2026.0680.79
COPSD (Hybrid)Base Model=Qwen3-VL-4B, Alignment Strategy=COPSD (Hybrid)2026.0680.67
COPSD (Standard)Base Model=Qwen3-VL-4B, Alignment Strategy=COPSD (Standard)2026.0680.57
VisionZero-RWModel Scale=4B2026.0680.24
LLaVA-NeXT-7BVisual Token Budget=2880, Token Reduction Ratio=100%, Base Model=LLaVA-NeXT-7B2026.0780.2
AnchorPruneVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.0780.2
BaseModel Scale=4B2026.0680.07
VanillaBackbone=LLaVA-1.5-13B, Token Budget=Full Tokens2026.0680
COPSD (Hybrid)Base Model=Qwen2.5-VL-7B, Alignment Strategy=COPSD (Hybrid)2026.0679.99
VisPlayModel Scale=4B2026.0679.89
Safe-RLHF-VBase Model=Qwen2.5-VL-7B, Alignment Strategy=Safe-RLHF-V2026.0679.73
Safe-RLHF-VBase Model=Qwen3-VL-4B, Alignment Strategy=Safe-RLHF-V2026.0679.65
GRPOBase Model=Qwen3-VL-4B, Alignment Strategy=GRPO2026.0679.51
VILA-UUnified=✓, #Params=7B, Visual Representation Type=Discrete2026.0679.4
AnchorPruneVisual Token Budget=320, Token Reduction Ratio=↓ 88.9%, Base Model=LLaVA-NeXT-7B2026.0779.4
SFTBase Model=Qwen3-VL-4B, Alignment Strategy=SFT2026.0679.35
LLaVA-1.5-7BModel Scale=7B-class, LLM=Vicuna-7B2026.0679.3
VisionZero-CLEVRModel Scale=2B2026.0679.04
DivPruneVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.0778.9
VisionZero-RWModel Scale=2B2026.0678.89
VisionZero-ChartModel Scale=2B2026.0678.85
Qwen-VLLLM=Qwen-7B2026.0678.8
COPSD (Standard)Base Model=Qwen2.5-VL-7B, Alignment Strategy=COPSD (Standard)2026.0678.75
8-shot ICLBackbone=LLaVA-OneVision (Qwen2-7B)2026.0678.7
PyramidDropVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.0778.7
AnchorPruneVisual Token Budget=160, Token Reduction Ratio=↓ 94.4%, Base Model=LLaVA-NeXT-7B2026.0778.6
VisPlayModel Scale=2B2026.0678.59
VISEModel Scale=2B2026.0678.54
STSBackbone=LLaVA-1.5-13B, Token Budget=128 Tokens2026.0678.5
VanillaBackbone=LLaVA-1.5-7B, Retained Tokens=576 (Upper Bound)2026.0678.5
LLaVA-1.5-7BBackbone=LLaVA-1.5-7B, Retained Tokens=5762026.0678.5
LLaVA-1.5LLM=Vicuna-7B2026.0678.5
LLaVA-1.5-7BRetain Tokens=576, Pruning Ratio=0%2026.0778.5
BaseModel Scale=2B2026.0678.37
LLaVA-1.5 + PADL+ (8×)LLM=Vicuna-7B, Compression Method=PADL, Compression Ratio=8x, Adapter fine-tuning=true2026.0678.2
Qwen-VL-ChatModel Scale=7B-class, LLM=Qwen-7B2026.0678.2
CDPrunerVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.0778.2
iReasonerModel Scale=2B2026.0677.94
EvoLMMModel Scale=2B2026.0677.86
Zoo-PruneBackbone=LLaVA-1.5-13B, Token Budget=128 Tokens2026.0677.8
OPDBase Model=Qwen2.5-VL-7B, Alignment Strategy=OPD2026.0677.62
LLaVA-1.5 + FreePrunerLLM=Vicuna-7B, Compression Method=FreePruner2026.0677.6
AgilePruneBackbone=LLaVA-1.5-13B, Token Budget=128 Tokens2026.0677.5
ShikraLLM=Vicuna-13B2026.0677.4
STSBackbone=LLaVA-1.5-7B, Retained Tokens=1282026.0677.3
VisionZipVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.0777.3
STSBackbone=LLaVA-1.5-13B, Token Budget=64 Tokens2026.0677.2
FastVVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.0777.2
DivPruneBackbone=LLaVA-1.5-13B, Token Budget=128 Tokens2026.0677.1
SparseVLMVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.0777.1
LLaVA-1.5 + PADL (8×)LLM=Vicuna-7B, Compression Method=PADL, Compression Ratio=8x, Adapter fine-tuning=false2026.0676.9
VanillaVision Token=576, Backbone=Qwen2.5-3B, Vision Encoder=clip-vit-large-patch14-3362026.0676.9
VisionZipBackbone=LLaVA-1.5-13B, Token Budget=128 Tokens2026.0676.8
LiquidUnified=✓, #Params=7B, Visual Representation Type=Discrete2026.0676.8
UniTokUnified=✓, #Params=7B, Visual Representation Type=Discrete2026.0676.8
CDPrunerVisual Token Budget=320, Token Reduction Ratio=↓ 88.9%, Base Model=LLaVA-NeXT-7B2026.0776.8
SPAREBackbone=LLaVA-1.5-7B, Retained Tokens=1282026.0676.7
EADPRetain Tokens=128, Pruning Ratio=77.8%2026.0776.7
LLaVA-1.5-7BBackbone=LLaVA-1.5-7B, Visual Token Budget=5762026.0776.7
Zoo-PruneBackbone=LLaVA-1.5-7B, Retained Tokens=1282026.0676.6
CDPrunerBackbone=LLaVA-1.5-7B, Retained Tokens=1282026.0676.6
LLaVA-1.5 + PruMergeLLM=Vicuna-7B, Compression Method=PruMerge2026.0676.6
DivPruneVisual Token Budget=320, Token Reduction Ratio=↓ 88.9%, Base Model=LLaVA-NeXT-7B2026.0776.6
Hyper-ICLBackbone=LLaVA-Interleave-7B2026.0676.51
Zoo-PruneBackbone=LLaVA-1.5-13B, Token Budget=64 Tokens2026.0676.4
AgilePruneBackbone=LLaVA-1.5-7B, Retained Tokens=1282026.0676.4