ResearchBenchmarksMathematical Reasoning on (AIME24, MATH500, GSM8k)Follow79.8Average AccuracyGRPO62.74467.17271.676.028May 18, 2026Evaluation ResultsMethodMethodLinksAverage AccuracyGRPOBackbone=Qwen3-4BBackbone=Qwen3-4B2026.0579.8CorR-POBackbone=Qwen3-4BBackbone=Qwen3-4B2026.0579.8DAPOBackbone=Qwen3-4BBackbone=Qwen3-4B2026.0579.5GSPOBackbone=Qwen3-4BBackbone=Qwen3-4B2026.0578.3Dr.GRPOBackbone=Qwen3-4BBackbone=Qwen3-4B2026.0574.4BaseBackbone=Qwen3-4BBackbone=Qwen3-4B2026.0563.4