List your company

Scientific Reasoning on GPQA (Accuracy, Inference Latency (s))

69.89AccuracyHDPO
43.411650.285857.1664.0342Jun 2, 2026

Evaluation results

MethodLinks
HDPOBackbone=Qwen2.5-Math-7B2026.0669.890.28
Critique-GRPOBackbone=Qwen2.5-Math-7B2026.0667.280.33
LUFFYBackbone=Qwen2.5-Math-7B2026.0661.510.24
GRPOBackbone=Qwen2.5-Math-7B2026.0660.180.17
SimpleRLBackbone=Qwen2.5-Math-7B2026.0647.310.13
EDGE-GRPOBackbone=Qwen2.5-Math-7B2026.0646.760.14
OatBackbone=Qwen2.5-Math-7B2026.0644.430.13