List your company

Multimodal Reasoning on BLINK

69.86AccuracyGPT-5
14.386428.788243.1957.5918Mar 18, 2025Jun 2, 2025Aug 17, 2025Nov 1, 2025Jan 16, 2026Apr 2, 2026Jun 17, 2026

Evaluation results

MethodLinks
GPT-53-run average=true2026.0669.86—
Qwen3-VL-8B3-run average=true2026.0669.43—
Qwen3-VL-32B3-run average=true2026.0666.22—
Qwen3-VL-8B-InstructParameters=8B, Instruct=true2026.0563.3—
InternVL3.5-38B3-run average=true2026.0662.65—
CAVE-7BParameters=7B2026.0561.5—
Visual-OPSD3-run average=true, Latency (s)=10.02026.0661.44—
GPT-4o3-run average=true2026.0660.28—
InternVL3.5-8BParameters=8B2026.0559.8—
InternVL3.5-8B3-run average=true2026.0659.6—
ThinkMorph3-run average=true, Latency (s)=142.82026.0659.49—
Qwen2.5-VLParameters=7B2026.0558.5—
PulseFocus (Qwen3-VL)Params=2B, Configuration=PulseFocus, Budget control=true2026.0356.40.85
DeepEyes-7BParameters=7B2026.0556.4—
LLaVA-NeXTSOFA=true2025.0355.92—
Qwen3-VLParams=2B, Configuration=Baseline2026.0355.55—
Visual-OPSD-Noise3-run average=true, Latency (s)=14.92026.0655.49—
Text-only SFT3-run average=true, Latency (s)=28.52026.0654.39—
PulseFocus (InternVL3.5)Params=8B, Configuration=PulseFocus, Budget control=true2026.0354.183.73
LLaVA-NeXTSOFA=false2025.0353.34—
MantisSOFA=true2025.0353.19—
MantisSOFA=false2025.0351.41—
InternVL3.5Params=8B, Configuration=Baseline2026.0350.45—
VILASOFA=true2025.0350.29—
VILASOFA=false2025.0349.79—
Idefics2SOFA=true2025.0348.31—
BAGEL-7B3-run average=true2026.0647.66—
Idefics2SOFA=false2025.0345.78—
InternVL2SOFA=true2025.0343.26—
Random Guessing2025.0339.27—
InternVL2SOFA=false2025.0338.95—
Janus-Pro-7B3-run average=true2026.0638.51—
Chameleon-7B3-run average=true2026.0616.52—