ResearchBenchmarksMathematical Reasoning on Minerva Math (avg@4, Average)Follow52.9Accuracy (avg@4)DC-GRPO49.98850.74451.552.256Jun 7, 2026Evaluation ResultsMethodMethodLinksAccuracy (avg@4)Average AccuracyDC-GRPOBackbone=Qwen3-30B-A3B...Backbone=Qwen3-30B-A3B, Precision=BF162026.0652.971.8GRPOBackbone=Qwen3-30B-A3B...Backbone=Qwen3-30B-A3B, Precision=BF162026.065269GRPOBackbone=Qwen3-8B, Pre...Backbone=Qwen3-8B, Precision=BF162026.065167.3DC-GRPOBackbone=Qwen3-8B, Pre...Backbone=Qwen3-8B, Precision=BF162026.0650.171