Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Visual logic on VlmsAreBlind
Loading...
66.79
Top-1 Accuracy
Qwen3-VL-8B-Instruct (Teacher)
48.3196
53.1148
57.91
62.7052
May 30, 2026
Top-1 Accuracy
Top-16 Accuracy
Updated 1d ago
Evaluation Results
Method
Method
Links
Top-1 Accuracy
Top-16 Accuracy
Qwen3-VL-8B-Instruct (Teacher)
Model Size=8B
2026.05
66.79
-
VGS On-Policy Distillation
Student Model=Qwen3-VL...
2026.05
65.49
65.92
Standard On-Policy Distillation
Student Model=Qwen3-VL...
2026.05
64.46
64.82
Qwen3-VL-4B-Instruct (Initial Student)
Model Size=4B
2026.05
61.94
-
VGS On-Policy Distillation
Student Model=Qwen3-VL...
2026.05
54.11
53.24
Standard On-Policy Distillation
Student Model=Qwen3-VL...
2026.05
51.86
50.52
Qwen3-VL-2B-Instruct (Initial Student)
Model Size=2B
2026.05
49.03
-
Feedback
Search any
task
Search any
task