ResearchBenchmarksMathematical Reasoning on HMMT25 (Acc avg@32)Follow82.2Accuracy avg@32IOP-GSPO50.27258.56166.8575.139Apr 19, 2026Evaluation ResultsMethodMethodLinksAccuracy avg@32IOP-GSPOModel Architecture=Qwe...Model Architecture=Qwen3-Next2026.0482.2GSPOModel Architecture=Qwe...Model Architecture=Qwen3-Next2026.0475.4BaseModel Architecture=Qwe...Model Architecture=Qwen3-Next2026.0473.9IOP-GSPOModel Architecture=Qwe...Model Architecture=Qwen3-32B2026.0463.2GSPOModel Architecture=Qwe...Model Architecture=Qwen3-32B2026.0455.4BaseModel Architecture=Qwe...Model Architecture=Qwen3-32B2026.0451.5