List your company

Visual Policy Reasoning on ClevrPolicy-M N=4

99.4AccuracyTriMPI w/ PoRo-DAPO
10.4833.56556.6579.735Oct 10, 2025

Evaluation results

MethodLinks
TriMPI w/ PoRo-DAPOComplexity=N=42025.1099.4
TriMPI w/ PoRo-GRPOComplexity=N=42025.1099.05
CoT SFT + DAPOComplexity=N=42025.1088.75
CoT SFT + GRPOComplexity=N=42025.1087.1
CoT SFTComplexity=N=42025.1055.35
Direct SFTComplexity=N=42025.1037.85
In-ContextComplexity=N=42025.1013.9