Last week, ByteDance's Seed large model department underwent a fresh round of organizational restructuring, marked by the introduction of four new primary departments. The Pretrain Data department, helmed by Li Chenggang, amalgamates formerly fragmented pre-training data teams specializing in text models, programming, visual comprehension, and voice. This unified entity is tasked with managing the multimodal data for the novel Omni model and undertaking the data-related tasks essential for the pre-training of ultra-large models. The Horizon RL department, under the guidance of Tang Shengyu, consolidates previously dispersed post-training teams focusing on post-training, inference, and visual understanding. Its primary responsibility lies in leveraging reinforcement learning to bolster the foundational intelligence capabilities of models. The Product Posttrain-Work department, led by Qin Yujia, primarily caters to B-end applications. It oversees the integration and deployment of Agentic models, refining their Agentic capabilities for office environments, including facilitating the task modes of Doubao and Dola. The Application team, previously overseen by Zhu Wenjia, has been rebranded as Product Posttrain-Chat. It continues to predominantly serve C-end applications, managing the integration and deployment of dialogue models, and collaborates closely with the Product Posttrain-Work department. All four department heads report to Wu Yonghui. Furthermore, Seed has designated leaders for AI safety and other pioneering exploration domains.