List your company

Reasoning on GPQA Diamond (P@1, P@32, #Tok)

28.41Precision@1TARPO
26.766827.193427.6228.0466Jun 4, 2026

Evaluation results

MethodLinks
TARPOBackbone=Qwen2.5-3B-Instruct2026.0628.4190.91568.3
Entropy-RoutedBackbone=Qwen2.5-3B-Instruct2026.0628.390.91373.2
GRPOBackbone=Qwen2.5-3B-Instruct2026.0628.0383.33608.5
HRPOBackbone=Qwen2.5-3B-Instruct2026.0627.5188.38603.3
Pure LatentBackbone=Qwen2.5-3B-Instruct2026.0627.2637.37405.6
CoTBackbone=Qwen2.5-3B-Instruct2026.0626.8392.93410.6