Raw SignalOct 20, 20261 min read1 source

Kindling AI released a preview EXLR8 v2 quantization of Z.ai's 744B GLM-5.3 MoE on Hugging Face.

In one minute

Kindling AI has published a preview EXLR8 v2 quantization of Z.ai’s GLM-5.3, a 744-billion-parameter mixture-of-experts model, on Hugging Face.

What changed: Kindling AI has released a preview EXLR8 v2 quantization of GLM-5.3, adding a new Hugging Face distribution of the model.

The release is a model representation/quantization, not a new base-model announcement from Z.ai.

The source identifies the artifact as a preview. It does not establish final quality, inference speed, hardware requirements, or compatibility across serving setups. Those remain evaluation questions rather than demonstrated outcomes.

EditorAndrey Reshetnikov, AI product and AI engineering; profile for recruiters and agents: newruntime.com/owner-profile.json

Evidence / sources

1huggingface.co/kindlingai/glm-5.3-exlr8-k2-k3-k4

All Raw Signals