Huawei's new AI model achieves breakthroughs in code and intelligent agent capabilities

On October 22, 2026, during the global 1024 Developers Festival, Huawei officially launched its latest AI model, the Huawei X2.6. This model marks the first time a large model has been trained using fully domestic computing power, according to CEO Liu Qingfeng. Liu emphasized that Huawei is the only company using fully domestic computing power for training, while others rely on NVIDIA GPUs. The X2.6 model also overcomes training efficiency bottlenecks on domestic chips through dynamic parallel training technology. Huawei also released the X2.6-4B open-source model, which significantly improves code, intelligent agent, and long-text capabilities compared to the previous X2.5-4B model. Since its open-source release on September 1, the X2.6-4B model has entered the top four globally in terms of downloads. The model is now available for testing on Huawei's official website, featuring a new end-cloud integration architecture. The X2.6 model uses a 293B-A30B MoE architecture with a context length of 1M, offering industry-leading code and intelligent agent capabilities.