[Zhipu Launches and Open-Sources GLM-5.3-Flash] Odaily Planet Daily reports that Zhipu officially announced the launch and open-sourcing of GLM-5.3-Flash, the first native multimodal model in the GLM-5 series. According to reports, the overall performance of GLM-5.3-Flash surpasses GLM-5.2, with programming and Agent evaluations approaching Claude Opus 4.8, while its cost is only one-tenth of GLM-5.2. Additionally, it features a new foundational model, introducing sparse attention and linear attention hybrid architecture to the main GLM series for the first time, and is pre-trained with 30T Token multimodal data.