TY  - RPRT
TI  - Revisiting the 16x Pretraining Compute Gap between Diffusion and Autoregressive Language Models
AU  - Guanghan Wang
AU  - Bing He
AU  - Zhan Shi
AU  - Benoit Dumoulin
AU  - Hanqing Lu
AU  - Volodymyr Kuleshov
PY  - 2026
UR  - https://arxiv.org/abs/2610.17590
ID  - 2610.17590
ER  -