List your company
ResearchFormerly WizWand

Visual Perception and Reasoning on BLINK

85.69AccuracyOurs-8B
37.756450.200762.64575.0893Feb 24, 2026Mar 13, 2026Mar 30, 2026Apr 16, 2026May 3, 2026May 20, 2026Jun 7, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Ours-8BParameters=8B2026.0685.69
Qwen3-VL-8BParameters=8B2026.0684.08
RoboBrain2.0-7BParameters=7B2026.0683.95
CGC-8BBase MLLM=Qwen3-VL2026.0470.02
Qwen3-VL-8B2026.0468.33
Embodied-R1-7BParameters=7B2026.0666.73
InternVLA-M1Parameters=M12026.0665.65
GPT-4o2026.0263
InternVL3-8BParameters=8B2026.0661.36
MiCo-7BBase MLLM=Qwen2.5-VL2026.0457.23
HyLaR-7BModel Type=Visual-Latent Model, Number of Parameters=7B2026.0457.14
LaserModel Type=Visual-Latent Model2026.0456.92
SKILA2026.0256.7
LIVR2026.0256.3
CGC-7BBase MLLM=Qwen2.5-VL2026.0456.23
ThymeModel Type=Thinking-with-Images Agent Model2026.0456.1
NoisyRollout-7BBase MLLM=Qwen2.5-VL2026.0456.07
CoVT2026.0256
CrystaLBase Model=Qwen2.5-VL-7B2026.0255.9
InternVL3-8B2026.0455.81
ThinkLite-VL-7BBase MLLM=Qwen2.5-VL2026.0455.81
Qwen2.5-VL-7BParameters=7B2026.0255.7
ZoomEyeModel Type=Thinking-with-Images Agent Model2026.0455.55
Qwen2.5-VL-7B2026.0455.54
InternVL3.5-8BModel Type=Open-Source Model, Number of Parameters=8B2026.0454.81
VLAA-Thinker-7BBase MLLM=Qwen2.5-VL2026.0454.81
HyLaR-SFTModel Type=Visual-Latent Model, Training Protocol=SFT2026.0454.5
Perceval#Param=7B2026.0454.49
MM-Eureka-7BBase MLLM=Qwen2.5-VL2026.0454.39
Qwen2.5-VL-7BModel Type=Open-Source Model, Number of Parameters=7B2026.0453.6
LVRModel Type=Visual-Latent Model2026.0453.6
Qwen2.5-VL + GRPO#Param=7B2026.0453.55
LVR2026.0252.5
Qwen2-VL-7B2026.0452.4
Migician-7BBase MLLM=Qwen2-VL2026.0451.53
OpenVLThinker#Param=7B2026.0451.36
Pixel-Reasoner#Param=7B, Tool Capability=true2026.0451.1
DeepEyesModel Type=Thinking-with-Images Agent Model2026.0451.08
Vision-R12026.0251
DeepEyes#Param=7B, Tool Capability=true2026.0450.98
MonetModel Type=Visual-Latent Model2026.0450.71
MM-Eureka#Param=7B2026.0450.23
VL-Rethinker#Param=7B2026.0449.91
Vision-R1#Param=7B2026.0449.72
VLAA-Thinker#Param=7B2026.0449.38
LLaVA-OneVisionModel Type=Open-Source Model2026.0449.34
Qwen2.5-VL + GRPO#Param=3B2026.0449.13
IC-VCOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.0549.01
IC-VCOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.0548.92
Perceval#Param=3B2026.0448.75
Qwen2.5-VL#Param=7B2026.0448.56
S-VCOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.0548.31
mDPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.0547.99
SymMPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.0547.89
DPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.0547.75
S-VCOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.0547.41
mDPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.0547.25
DPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.0547.2
R1-VL#Param=7B2026.0447.19
V-DPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.0547.09
Qwen2.5-VL#Param=3B2026.0446.94
SymMPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Synthetic2026.0546.83
V-DPOBackbone=LLaVA-OneVision-Qwen2-7B, Contrastive Sample Source=Contrastive Sample Editing2026.0546.63
LMM-R1#Param=3B2026.0446.6
Perception-R1#Param=3B2026.0446.44
VLM-R1#Param=3B2026.0446.25
Jigsaw-R1#Param=3B2026.0445.01
LLaVA-OneVision-Qwen2-7BBackbone=LLaVA-OneVision-Qwen2-7B2026.0544.82
R1-VL#Param=2B2026.0442.81
Claude-4-Sonnet2026.0239.6