ByteDance has postponed the release of Doubao large model 2.2, which was originally scheduled for August. The company is extending both pretraining and post-training for its Seed model, with added emphasis on coding, tool calling and agent capabilities. People familiar with ByteDance told Whale Lab that the target is to place the model’s coding performance in the industry’s top tier before the end of the year, benchmarking it against Zhipu GLM-5.2 and Kimi K3.