ResearchBenchmarksMath Reasoning on AIME 25 (pass@1, cons@32)Follow35.72Pass@1XRPO25.652828.266430.8833.4936Oct 8, 2025Evaluation ResultsMethodMethodLinksPass@1Consistency @32XRPOBackbone=Qwen3-1.7BBackbone=Qwen3-1.7B2025.1035.7250GSPOBackbone=Qwen3-1.7BBackbone=Qwen3-1.7B2025.1033.2346.67DAPOBackbone=Qwen3-1.7BBackbone=Qwen3-1.7B2025.1032.533.33TPO-SBackbone=Qwen3-1.7BBackbone=Qwen3-1.7B2025.1026.0433.33