List your company

Reasoning on ARC Challenge (Pass@1, Pass@32)

87.2Pass@1LEPO
63.602469.728775.85581.9813Apr 20, 2026

Evaluation results

MethodLinks
LEPOBackbone=Qwen2.5-7B2026.0487.298.89
GRPOBackbone=Qwen2.5-7B, Latent Inference=false2026.0487.0898.98
GRPO (w/ latent inference)Backbone=Qwen2.5-7B, Latent Inference=true2026.0487.0798.72
Soft TokensBackbone=Qwen2.5-7B2026.0486.6795.48
HRPOBackbone=Qwen2.5-7B2026.0486.1294.71
CoTBackbone=Qwen2.5-7B, Latent Inference=false2026.0465.3499.91
CoT (w/ latent inference)Backbone=Qwen2.5-7B, Latent Inference=true2026.0464.5199.91