List your company

General Science Reasoning on GPQA Diamond (Accuracy, Length)

57.68AccuracyDDPO
50.753652.551854.3556.1482Mar 19, 2026

Evaluation results

MethodLinks
DDPOBackbone=Qwen3-14B2026.0357.685,683
L1Backbone=Qwen3-14B2026.0357.265,543
GRPOBackbone=Qwen3-14B2026.0357.115,728
SHORTER BETTERBackbone=Qwen3-14B2026.0356.884,416
GRPOBackbone=Qwen3-8B2026.0356.767,379
A-DLPBackbone=Qwen3-14B2026.0356.414,986
DDPOBackbone=Qwen3-8B2026.0356.356,363
GRPO-LEADBackbone=Qwen3-14B2026.0355.017,147
DDPOBackbone=Qwen3-4B2026.0351.846,252
GRPOBackbone=Qwen3-4B2026.0351.027,423