GLM-5.3 Flash
Zhipu AI (Z.ai) · 2026-08-26 · 321.3B parameters
GLM-5.3 Flash is Z.ai's first natively multimodal GLM-5 model, built from a new base with hybrid sparse and linear attention. It accepts text, image, and video inputs, supports low, high, and max reasoning effort, and releases its weights under MIT. The released language configuration supports 1,048,576 positions, while Z.ai reports 320B total and 18B activated parameters; the released checkpoint holds 321,323,019,006 parameters excluding FP8 scale tensors and router bias buffers.