zai-org

GLM-5.3

Z.ai's 743B-parameter (39B active) MoE model with 1M-token context for reasoning, coding and agents, shipped natively in FP8.text generationother2M downloadsHugging Face

No server in stock right now has the 893 GB this model needs for inference.

License restrictions

Commercial use allowed; Model-as-a-Service operators with over USD 10B combined revenue in 12 months must pass Z.ai's security review.

GPU memory it needs

From model cards and vendor documentation. Lower precision needs less memory, with slightly lower quality. Estimates are marked.

What you want to do	Precision	Minimum	Recommended	Setup	Basis
Run the model vLLM recipe minimum for this repo's native FP8 checkpoint with FP8 KV cache; the recipe's default profile is 8x B300.	FP8, 8-bit	893 GB	1128 GB	8x H200 (vLLM recipe, single-node)	Published
Run the model vLLM recipe minimum for separate repo Inferact/GLM-5.3-NVFP4 (NVFP4 experts, rest FP8/BF16), Blackwell only.	4-bit	558 GB	—	Blackwell (B200/B300)	Published