ResearchBenchmarksMathematical Reasoning on HMMT (Acc@16)Follow34.6Accuracy@16AMR-SD27.3229.2131.132.99May 18, 2026Evaluation ResultsMethodMethodLinksAccuracy@16AMR-SDBackbone=Qwen3-8B, Thi...Backbone=Qwen3-8B, Thinking Mode=true2026.0534.6GRPOBackbone=Qwen3-8B, Thi...Backbone=Qwen3-8B, Thinking Mode=true2026.0531.7Qwen3-8BThinking Mode=trueThinking Mode=true2026.0528.3RLSDBackbone=Qwen3-8B, Thi...Backbone=Qwen3-8B, Thinking Mode=true2026.0527.6