ResearchBenchmarksMathematical Reasoning on AIME 25 (Acc avg@32)Follow92.1Accuracy avg@32IOP-GSPO72.13277.31682.587.684Apr 19, 2026Evaluation ResultsMethodMethodLinksAccuracy avg@32IOP-GSPOModel Architecture=Qwe...Model Architecture=Qwen3-Next2026.0492.1GSPOModel Architecture=Qwe...Model Architecture=Qwen3-Next2026.0488.4BaseModel Architecture=Qwe...Model Architecture=Qwen3-Next2026.0487.8IOP-GSPOModel Architecture=Qwe...Model Architecture=Qwen3-32B2026.0483.5GSPOModel Architecture=Qwe...Model Architecture=Qwen3-32B2026.0476.6BaseModel Architecture=Qwe...Model Architecture=Qwen3-32B2026.0472.9