Reports say Doubao 2.2 has been delayed: ByteDance catches up on coding, recruitment explicitly names Claude Code and Codex

Beating AI Insight News Brief: ByteDance’s originally planned August launch of Doubao 2.2 has been delayed. The company will allocate more training time to focus on strengthening coding, tool calling, and agent capabilities. This year, one of Seed’s core goals is to elevate its coding models to the top tier. Internally, the team aims to achieve performance on par with GLM-5.2 and Kimi-K3 by the end of the year, to earn genuine developer recognition for ByteDance’s coding models. In August, Seed underwent a major restructuring: teams originally segmented by text, speech, code, and vision were reorganized into four departments: Pretrain Data, Horizon RL, Product Posttrain-Work, and Product Posttrain-Chat. Horizon RL will handle coding post-training, while the Work team will focus on tool calling, GUI operations, and long-task execution. ByteDance has recently been intensively recruiting for these areas: its official website is hiring Code Agents, general Agents, and reinforcement learning algorithm engineers. One Multi-Agent Harness position explicitly requires research on Coding Agents such as Claude Code and Codex, plus building multi-agent and reinforcement learning environments capable of sustained long runs. When Seed 2.1 launched in June, it already highlighted coding, but ByteDance clearly deemed it inadequate. This time, Doubao 2.2 will delay its launch to first refine its coding capabilities to a more robust standard.