zai-org/GLM-5.3 model card

Card read 2026-10-17. States GLM-5.3 uses the same base model as GLM-5.2 and that all gains are post-training. Model overview table: 744B total parameters, 40B active parameters, 78 layers, 256 experts with 8 activated, MLA + DSA, context 1M, quantized weight size about 375GB. Deployment section names SGLang, vLLM, xLLM (Ascend), and KTransformers, and says llama.cpp is not supported yet. HF API createdAt 2026-08-25T06:42:50Z; safetensors total 753,329,940,480 (BF16 2.10B, F8_E4M3 751.23B, F32 19,456). The 40B active figure is a vendor claim; the 753.33B total is the safetensors count.