List your company

Reasoning on GPQA Diamond (Pass@1, Pass@32)

29.96Pass@1 (Reasoning)LEPO
9.877615.091320.30525.5187Apr 20, 2026Apr 23, 2026Apr 27, 2026Apr 30, 2026May 4, 2026May 7, 2026May 11, 2026

Evaluation results

MethodLinks
LEPOBackbone=Qwen2.5-7B2026.0429.9690.91
Soft TokensBackbone=Qwen2.5-7B2026.0428.6981.31
GRPOBackbone=Qwen2.5-7B, Latent Inference=false2026.0428.2887.88
GRPO (w/ latent inference)Backbone=Qwen2.5-7B, Latent Inference=true2026.0428.0988.89
HRPOBackbone=Qwen2.5-7B2026.0427.8478.79
FOCUSFTn=322026.0519.480.8
Standard SFTn=322026.0515.678.8
CoTBackbone=Qwen2.5-7B, Latent Inference=false2026.0410.9282.83
CoT (w/ latent inference)Backbone=Qwen2.5-7B, Latent Inference=true2026.0410.6583.33