LLaDA2.0: Scaling Up Diffusion Language Models to 100B

Benchmark Model Rank Results
math-word-problem-solving-on-mathLLaDA2.0-flash#1Accuracy: 95.44
mathematical-reasoning-on-olympiadbenchLLaDA2.0-flash#3Accuracy: 74.07
mathematical-reasoning-on-omni-mathLLaDA2.0-flash#3Accuracy: 50.3
logic-puzzle-solving-on-zebralogicLLaDA2.0-flash#5Accuracy: 82.3
code-generation-on-multipl-eLLaDA2.0-flash#6Accuracy: 74.87
arithmetic-reasoning-on-gsm8kLLaDA2.0-flash#11Accuracy: 96.06
code-generation-on-mbppLLaDA2.0-flash#12Accuracy: 88.29
code-generation-on-humanevalLLaDA2.0-flash#17Pass@1: 94.51
multi-task-language-understanding-on-c-evalLLaDA2.0-flash#17Accuracy: 86.75
sentence-completion-on-hellaswagLLaDA2.0-flash#39Accuracy: 84.97
multi-task-language-understanding-on-mmlu-proLLaDA2.0-flash#61Accuracy: 73.36
mathematical-reasoning-on-aime-2025LLaDA2.0-flash#71Accuracy: 60.0