ResearchBenchmarksMathematical Reasoning and Code Generation on Average (GSM8K, MATH, HumanEval, MBPP)Mathematical Reasoning and Code Generation on Average (GSM8K, MATH, HumanEval, MBPP)Follow52.06AccuracyFeF-DLLM38.789642.234845.6849.1252May 14, 2026Evaluation resultsMethodMethodLinksAccuracySpeed FactorFeF-DLLMdiffusion model steps=4diffusion model steps=42026.0552.062.33FeF-DLLMdiffusion model steps=2diffusion model steps=22026.0551.793.86SSD2026.0548.582.09DCD2026.0548.050.43LLaDAtokens per step=1tokens per step=12026.0546.751LLaDA/2tokens per step=2tokens per step=22026.0542.31.98DDOSP2026.0539.31.98