[Doubao 2.2 Delayed: ByteDance Focuses on Coding, Directly Targets Claude Code and Codex in Recruitment] Breaking AI News from Dongcha Beating AI: ByteDance's originally planned August release of Doubao 2.2 has been postponed. ByteDance intends to spend more time on training, focusing on coding, tool invocation, and agent capabilities. One of the goals set for Seed this year is to elevate its coding capabilities to the top tier. Internally, the company hopes to achieve results similar to GLM-5.2 and Kimi-K3 by the end of the year, enabling developers to truly recognize ByteDance's coding model. In August, Seed underwent a major restructuring. The teams originally divided by text, voice, code, and vision were reorganized into four departments: Pretrain Data, Horizon RL, Product Posttrain-Work, and Product Posttrain-Chat. Horizon RL will be responsible for post-training in coding, while the Work team will focus on tool invocation, GUI operations, and executing long tasks. ByteDance has also been actively recruiting talent in these areas recently. The official website is currently hiring for positions such as Code Agent, General Agent, and Reinforcement Learning Algorithm Engineer. One job listing for a Multi-Agent Harness position explicitly states that the role involves researching coding agents like Claude Code and Codex, as well as building multi-agent and reinforcement learning environments capable of running continuously for extended periods. When Seed 2.1 was released in June, it already emphasized coding, but ByteDance clearly feels that this is still insufficient. For Doubao 2.2, the company is willing to delay the release to ensure that its coding capabilities are significantly improved. [Original Link]