Zhipu AI Launches and Open-Sources GLM-5.3-Flash Model

Zhipu AI has officially released and open-sourced GLM-5.3-Flash, the first native multimodal model in the GLM-5 series. The model utilizes a hybrid architecture of sparse and linear attention, trained on 30T tokens of multimodal data. It offers performance comparable to Claude Opus 4.8 in programming and agent tasks while being significantly more cost-effective than its predecessor, GLM-5.2. Previously tested anonymously as Ox-Alpha on platforms like OpenCode and OpenRouter, the model demonstrated strong performance and high demand. Developers can now deplo