ResearchBenchmarksMathematical Reasoning on AIME25 (Accuracy)Follow76.3AccuracyLUSPO0.69220.32139.9559.579Feb 5, 2026Evaluation ResultsMethodMethodLinksAccuracyLUSPOBase model=Qwen3-30B-A...Base model=Qwen3-30B-A3B-Instruct2026.0276.3GSPOBase model=Qwen3-30B-A...Base model=Qwen3-30B-A3B-Instruct2026.0259.2w/o RLVRBase model=Qwen3-30B-A...Base model=Qwen3-30B-A3B-Instruct2026.0257.2LUSPOBase model=Qwen2.5-7B-...Base model=Qwen2.5-7B-Base2026.0213.9GSPOBase model=Qwen2.5-7B-...Base model=Qwen2.5-7B-Base2026.0211.2w/o RLVRBase model=Qwen2.5-7B-...Base model=Qwen2.5-7B-Base2026.023.6