ResearchBenchmarksMathematical Reasoning on MATH 500 (pass@1 and pass@32)Follow92.16Pass@1FG-ExPO72.691277.745682.887.8544May 12, 2026Evaluation ResultsMethodMethodLinksPass@1Pass@32FG-ExPOBackbone=Qwen3-8B-BaseBackbone=Qwen3-8B-Base2026.0592.1696.8GRPOBackbone=Qwen3-8B-BaseBackbone=Qwen3-8B-Base2026.0591.7896.8FG-ExPOBackbone=DeepSeek-R1-D...Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0584.0396GRPOBackbone=DeepSeek-R1-D...Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0582.8895BaseBackbone=DeepSeek-R1-D...Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0573.4493.6