ByteDance’s Doubao AI launches “deep thinking” reasoning mode using purely self-developed algorithms

On March 6, ByteDance’s large-scale AI model Beanbao inference mode was launched, covering Q&A, search, writing and reading scenarios. You can experience it by turning on the “deep thinking” switch in the input box. Previously, DeepSeek, kimi, Tencent Yuanbao, and Wenxin Model have all launched their own inference models.

According to previous disclosures from the Doubao technical team, this function is based on the research and development of the Doubao 1.5Pro base model. It is a breakthrough through a purely self-developed reinforcement learning (RL) algorithm. It can complete training optimization without relying on external model data, achieving a balance between computing power efficiency and model performance.

On March 3, Tencent Yuanbao rose to the top of the free App download rankings in the Apple App Store in China, surpassing DeepSeek, and Doubao ranked fourth.