GLM-5.3-Flash

GLM-5.3-Flash is the first natively multimodal model in the GLM-5 series, with 320B total and 18B active parameters and a hybrid sparse-attention plus linear-attention architecture that cuts attention compute by 3.01x and KV cache size by 4.44x relative to GLM-5.3. It understands images, video, and documents natively, observes interfaces and rendered output during coding loops, and carries tasks across code, browsers, and graphical interfaces. It also handles professional deliverables such as PPTX, PDF, DOCX, and XLSX. The model accepts text, image, video, and file inputs and returns text, with a 1M-token context window and up to 128K output tokens. Thinking mode only supports enabled. The original prices shown here are the official CNY list prices converted at 7.1 CNY per USD: CNY 0.80 per 1M uncached input tokens, CNY 0.23 per 1M cache-hit input tokens, and CNY 2.80 per 1M output tokens. Cache storage is currently free for a limited time. The faster GLM-5.3-FlashX variant is exposed under the separate glm-5.3-flashx model name.