ResearchBenchmarksMathematical Reasoning on MATH 500 (Majority@2048)Follow78.8Majority@2048CurveRL69.64872.02474.476.776May 23, 2026Evaluation ResultsMethodMethodLinksMajority@2048CurveRLBackbone=Qwen3-1.7B-BaseBackbone=Qwen3-1.7B-Base2026.0578.8MaxRLBackbone=Qwen3-1.7B-BaseBackbone=Qwen3-1.7B-Base2026.0577.2GRPOBackbone=Qwen3-1.7B-BaseBackbone=Qwen3-1.7B-Base2026.0574.6BaseBackbone=Qwen3-1.7B-BaseBackbone=Qwen3-1.7B-Base2026.0570