ResearchBenchmarksMathematical Reasoning on AIME24, AIME25, and AMC23 (test)Follow86.67AIME24 pass@32CIPO58.933266.134173.33580.5359May 14, 2026Evaluation ResultsMethodMethodLinksAIME24 pass@32AIME25 pass@32AMC23 pass@32Average Pass@32CIPOBackbone=Qwen3-4BBackbone=Qwen3-4B2026.0586.677010085.56GRPOBackbone=Qwen3-4B, BS=128Backbone=Qwen3-4B, BS=1282026.0576.6763.339578.33GRPOBackbone=Qwen3-4B, BS=256Backbone=Qwen3-4B, BS=2562026.0573.33709579.44GRPOBackbone=Qwen3-4B, m=1...Backbone=Qwen3-4B, m=16, BS=1282026.0563.336092.571.94InitialBackbone=Qwen3-4BBackbone=Qwen3-4B2026.056053.3397.570.28