GLM-5.3

Checkpointzai-org/GLM-5.3Model card indexedModel card

Support summary

GLM-5.3 is a checkpoint published by Z.ai, built on the GlmMoeDsaForCausalLM architecture. vLLM 0.28.0, the newest indexed vLLM release, documents support. Official recipes or cookbooks cover SGLang, without pinning a stable release.

Documented hardware
Ascend NPU and NVIDIA / CUDA
Documented quantization
FP8
Documented features
Multi-token prediction, Reasoning parsers, Speculative decoding, Tensor parallelism, and Tool calling
Configured context
1,048,576 tokens
License
glm-5.3

Hardware, quantization and features count only statements about this checkpoint; engine-wide capabilities are listed with each release.

Hugging Face model card
Publisher
Z.ai
Task
text generation
License
glm-5.3
Library
transformers
Architecture
GlmMoeDsaForCausalLM
Weight tensor elements
753.3B
Tensor types
BF16, F8_E4M3, F32
Quantization
fp8
Configured context
1,048,576 tokens
Languages
en, zh
Weight access
Public
Repository created
2026-08-25
Source 1: zai-org/GLM-5.3 · publisher model cardSource 2: zai-org/GLM-5.3 · Hugging Face metadataSource 3: zai-org/GLM-5.3 · publisher configurationReviewed

Publisher deployment guidance

Model notes and sources

Engine support

3 engines

Engine support for GLM-5.3
EngineReleaseCheckpoint supportEvidence scopeLast known supported
vLLM0.28.0Newest indexedCheckpoint evidenceSource: vllm · models/zai-org/GLM-5.3.yamlReviewed 0.28.0This release
Features, dependencies and evidence for vLLM 0.28.0
SGLangVersion not pinnedRecipe / cookbookDocumented recipeSource 1: SGLang GLM-5.3 cookbook (documentation snapshot; runtime version not pinned)Source 2: SGLang GLM-5.3 hardware, precision and strategy recipesReviewed No supported release indexed
Features, dependencies and evidence for SGLang 0.5.19
TensorRT-LLM1.2.1Newest indexedNo matching evidence curatedNo supported release indexed
Features, dependencies and evidence for TensorRT-LLM 1.2.1

Recipes and cookbooks