On February 14, ByteDance officially announced that the Doubao Large Model has now advanced to Phase 2.0. This new iteration includes three versatile agent models—Pro, Lite, and Mini—alongside a dedicated Code model. The entire series has undergone meticulous optimization to meet the rigorous demands of large-scale production environments, with a particular focus on boosting multimodal comprehension, adeptly executing complex instructions, and handling long-term tasks with enhanced efficiency.
Among these models, the Pro version stands out for its prowess in deep reasoning and tackling long-chain tasks, positioning it as a formidable competitor to GPT 5.2 and Gemini 3 Pro. The Lite version strikes a perfect balance between performance and cost-effectiveness, outperforming the main model of the previous generation. The Mini version is crafted specifically for scenarios demanding low latency and high concurrency. Lastly, the Code version is tailored exclusively for programming tasks, delivering even superior results when used in conjunction with TRAE.