back

GLM-5

MIT

Z.ai · 744B (40B active) · Mixture of Experts

Frontier GLM MoE with 256 experts and strong agentic coding performance

HuggingFace
0 downloads 0 likes 2026-02 128K context

Use Cases

chat reasoning code

Mixture of Experts

Total experts: 256
Active experts: 8
Active params: 40.0B

Quantization Options

Quant Bits VRAM Quality Status
Q2_K 2 238.7 GB low —
Q3_K_M 3 334 GB moderate —
Q4_K_M 4 381.6 GB good —
Q5_K_M 5 476.9 GB good —
Q6_K 6 572.1 GB excellent —
Q8_0 8 762.7 GB excellent —
F16 16 1524.9 GB lossless —

About this model

HF model card GLM
GLM-5 source artwork

Z.ai

744B

Mixture of Experts model for chat, reasoning, code workloads.

Context

128K

Q4 VRAM

~381.6 GB

Image saved from public model sources

Mixture of Experts Thinking Tool use Code

GLM-5 on your hardware

Frontier GLM MoE with 256 experts and strong agentic coding performance. This page turns the Hugging Face model card into practical local-run numbers, so you can compare quantized VRAM, system RAM, and expected fit before downloading a large checkpoint.

Model shape

744B total parameters with 40B active. 8 of 256 experts are active per token.

Local fit target

Start with Q4_K_M: about 363.8 GB on disk and 381.6 GB VRAM before extra context and runtime overhead.

Best use cases

chat, reasoning, code

License and source

Listed as MIT from huggingface.co/zai-org/GLM-5.

Want the real verdict? Pick your GPU or edit the specs on this page and compare the quant table below.

Open HF repo