GLM-5.3

Name in the API

zai-org/GLM-5.3

Base model

zai-org/GLM-5.3

Weights we run

gpustack/GLM-5.3-W4A8 (W4A8 quantization: 4-bit weights, 8-bit activations)

Type

Chat

Input → output

text → text

Context length

524,288 tokens

Availability

HPC

Hardware

8 × NVIDIA H100 (94 GB) on two nodes of the Capella HPC cluster

Thinking

Default: on, with the highest effort (max).

reasoning_effort: every value is accepted, but the model only knows low and high. All other values (minimal, medium, xhigh, max) mean max. low and high both think clearly less than the default.

Numeric effort: no.

Turn off: not possible. With reasoning_effort: "none", chat_template_kwargs: {"enable_thinking": false} or chat_template_kwargs: {"thinking": false} the model still thinks, and the reasoning ends up in the answer (content) instead of the separate reasoning field (vLLM issue #54744).

Tool calling

Yes

Example

Chat request. Set model to the name above.

Notes

• For short answers, use reasoning_effort: "low" rather than turning thinking off, because turning it off mixes the reasoning into the answer.