Alpha
Contact
Implement · GLM

Can you run it?

Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input C ยท 65.2/100

This page is a projection of the one entry record, the Reliability and Data control factors that Implement covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.

Which domain expands which factor
  • AssessUse & modify + Transparency
  • ImplementData control + Reliability
  • UseReliability
  • SupportTransparency

Install & run

GLM checkpoints are standard safetensors on the verified zai-org org. The MoE flagships serve best on vLLM or SGLang, and the dense 0414 sizes also run under llama.cpp and Ollama. Always pin an exact revision.

# Hugging Face transformers (dense sizes).
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "zai-org/GLM-4-32B-0414"   # Pick the exact variant to deploy.
rev = "<pinned-commit-sha>"            # Pin the revision; do not float on main.
tok = AutoTokenizer.from_pretrained(model_id, revision=rev)
model = AutoModelForCausalLM.from_pretrained(model_id, revision=rev)
# vLLM (OpenAI-compatible server) - MoE flagships.
vllm serve zai-org/GLM-4.6 --revision <sha>

# SGLang - also first-class for the MoE.
python -m sglang.launch_server --model-path zai-org/GLM-4.6 --revision <sha>

# Ollama (community GGUF) - dense sizes.
ollama run glm4

The models are also available on Zhipu's ModelScope. Verify per-file checksums after download and confirm the LICENSE on the exact checkpoint.

Hardware & VRAM requirements

GLM-4-9B-0414 (dense) runs on a single modern GPU. GLM-4-32B-0414 (dense) needs a single high-memory GPU or dual GPU, and more at long context. GLM-4.5-Air (106B/12B MoE) needs a multi-GPU node, or aggressive quantization on a single large card. GLM-4.5 and GLM-4.6 (355B/32B MoE) need serious multi-GPU infrastructure, or heavy quantization, to serve at usable latency.

MoE total-parameter memory dominates even though only 32B are active per token, so size VRAM to the total and budget extra headroom for the KV cache at the 200K context of GLM-4.6.

Serving stacks

StackUse it for
vLLMHigh-throughput MoE serving and OpenAI-compatible API
SGLangFirst-class MoE serving (agentic/structured workloads)
llama.cppCPU, edge and GGUF for the dense 0414 sizes
OllamaLocal single-command runs (dense sizes)
transformersPrototyping and fine-tuning

vLLM and SGLang are the primary paths for the 355B and 106B MoE. Day-0 llama.cpp and edge support is strongest for the dense sizes and lags for the big MoE. This breadth underpins Operational Readiness (4), which stops short of 5 mainly on the infrastructure weight of the largest MoE.

Safe-deployment controls & Deployment Ceiling

The model-specific gate rows from the safe-deployment playbook are below.

  • Supply chain: pin the exact zai-org revision and verify checksums (match across Hugging Face and ModelScope). Confirm the per-checkpoint LICENSE and do not assume MIT everywhere.
  • Model layer: deploy the instruct checkpoints and set conservative decoding defaults.
  • Input/output layers: add input/output guardrails and your own guard classifier, because GLM ships no companion guard model. Add prompt-injection defences for agentic and coding paths and treat retrieved and tool content as untrusted.
  • Behaviour: test for China-aligned alignment and hosted-API-style moderation effects relevant to the use case.
  • Compliance: for the 355B MoE, resolve or avoid the EU systemic-risk / Article 55 question before EU high-stakes use, and record the AI-BOM posture.

Deployment Ceiling: T3 (bounded autonomous), conditional. With your own input/output guardrails, a guard model and prompt-injection defences, GLM's coding and agentic strength supports bounded-autonomous use. T3 additionally needs a deterministic action-policy engine and per-action human approval for irreversible operations. The release provides none of this external tooling.

Available quantizations

Community GGUF (llama.cpp, Ollama), AWQ/GPTQ and EXL quantizations circulate (bartowski, turboderp and others) and are best covered for the dense 0414 sizes. Coverage is partial because these are community efforts, with no broad first-class official quantization programme from Zhipu. Verify the source and check quality before relying on a third-party quant in production, especially for the MoE flagships.

Fine-tuning & adaptation

The MIT-licensed safetensors checkpoints support standard SFT and LoRA via transformers and TRL, and released inference code eases integration. The training data and training pipeline are not released, so continued pre-training or reproduction from scratch is not possible. The released weights can be adapted but not rebuilt. Placing a derivative on the EU market has EU AI Act provider implications (see Assess).

API / OpenAI-compatible integration

Serve GLM behind vLLM's or SGLang's OpenAI-compatible /v1 server for a drop-in integration with existing OpenAI-client code.

vllm serve zai-org/GLM-4.6 --revision <sha> --port 8000
# Then point any OpenAI SDK at http://localhost:8000/v1.

No model-specific client is required. The chat-completions surface behaves like any other OpenAI-compatible endpoint. Account for GLM's chat template and its agentic/thinking output format when parsing responses.

How this scores

The ownership factors this domain covers, drawn from the one entry record.

3

ReliabilityIs it reliable and good enough for the job?

Moderate

The models run dependably. Safety tuning is lighter than at Western frontier labs, there is no companion guard model, and the 355B MoE has an unresolved EU systemic-risk question.

How this scores (AOI sub-dimensions)
Operational4/5how practical it is to run, serve and maintain in productionEcosystem support is strong: dual distribution on Hugging Face and ModelScope, first-class vLLM and SGLang serving, community GGUF quants, and a size range from a 9B dense model to the 355B MoE.
Safety3/5whether misuse risks are evaluated and guardrails are providedInstruct variants are safety-tuned and withstand casual jailbreaks.
4

Doesn't extract your dataDoes running it keep your knowledge and data yours?

Strong

Self-hosted, the MIT-licensed weights run on the deployer's own infrastructure and nothing phones home, so data and derived knowledge stay with the deployer. The hosted Z.ai API is the only path that exposes inputs. It runs under a Singapore regime, distinct from the PRC: API content is 'not saved on our servers' and not used to train unless the user agrees, though consumer chat content is used to improve the models.

How this scores
Not a scored AOI dimension. For a self-hosted model, data-control is a structural property of running the weights yourself, strong by default unless the model phones home or the licence claws back rights. For a hosted API this factor is the retention + train-on-inputs + residency read, scored from the binding terms.
What this means for adoptionYou substantially but not uniformly own the GLM line, and the split is per variant. The current line is MIT: GLM-4.6 ships a checked-in MIT LICENSE ('Copyright (c) 2026 Zhipu AI') and the GLM-4.5 and GLM-4-0414 siblings declare the same MIT via HF metadata and README. They can be run, fine-tuned, distilled, redistributed and commercialised with no gate. The legacy glm-4-9b is not plain MIT. Its custom 'glm-4' licence requires commercial-use registration, 'Built with glm-4' attribution and a name prefix, and is governed by PRC law with disputes to Beijing's Haidian District court. Verify the LICENSE per checkpoint. Two things hold ownership at partial: training is opaque (closed data and code, China-aligned alignment that cannot be inspected) and no safety battery is benchmarked, so supply your own guards. The weights you self-host are yours. The hosted Z.ai service is Singapore-governed (SIAC arbitration), a different legal regime from the PRC glm-4-9b weights licence, and the two should be kept apart.

Sources

The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.

Model cardread2026-07-25
GLM-4.6 is a 355B-total / 32B-active MoE with a 200K context window, released by Zhipu AI / Z.ai and hosted on the zai-org Hugging Face org; the model card/metadata declares MIT.
Model cardread2026-07-25
GLM-4.5 (355B/32B) and GLM-4.5-Air (106B/12B) are MoE models on the zai-org Hugging Face org; the GLM-4.5 README states "They are released under the MIT open-source license and can be used commercially and for secondary development."
Model cardread2026-07-25
GLM checkpoints are hosted on the verified zai-org org on Hugging Face in safetensors with checksums (the current org; legacy checkpoints are under THUDM).
Licenceread2026-11-19
GLM-4.5, GLM-4.5-Air and GLM-4.6 declare the OSI-approved MIT license, which permits commercial use, via HF metadata ("license: mit") and README only.
Licenceread2026-11-19
GLM-5.2 ships an in-repo MIT LICENSE file: standard MIT text with no custom clauses, "Copyright (c) 2026 Zhipu AI." GLM-5.1 carries a byte-identical file and GLM-5.3-Flash a standard MIT file ("Copyright (c) 2026 Z.AI Co., Ltd").
Licenceread2026-11-19
Boundary evidence, outside the tracked checkpoints.
Licenceread2026-07-25
Licensing is per variant and the legacy glm-4-9b is NOT plain MIT.
Licenceread2026-11-19
The GLM-4-0414 family (dense GLM-4-32B-0414 and GLM-4-9B-0414, up to 128K extended context) was open-sourced under the MIT license.
Documentationunverified2026-07-25
Zhipu / Z.ai publishes GLM inference code and technical reports on GitHub (zai-org/GLM-4, zai-org/GLM-4.5), but not the training data or training pipeline.
Model cardunverified2026-07-25
GLM models are also officially published on Alibaba's ModelScope hub (dual distribution) by Zhipu.
Terms of serviceread2026-11-19
Hosted Z.ai Terms of Service (Last Update April 14, 2026): governed by Singapore law with SIAC arbitration (seat Singapore).
Privacy Policyread2026-11-19
Hosted Z.ai Privacy Policy (Last Update September 29, 2025): data is processed in Singapore.
Third-party analysisunverified2026-07-25
Third-party evaluation places GLM-4.6 among the stronger open models for real-world coding, tool use and agentic tasks.
Third-party analysisunverified2026-07-25
GLM models carry China-aligned alignment and the hosted Z.ai API applies Chinese content moderation on politically sensitive topics.
Third-party analysisunverified2026-07-25
GLM is supported for local serving on vLLM and SGLang (with community GGUF quants for the dense sizes) and distributed across Hugging Face and ModelScope.
Third-party analysisunverified2026-07-25
GLM instruct variants are safety-tuned but with lighter alignment coverage than Western frontier labs and no companion guard model.
Model cardunverified2026-07-25
No public EU AI Act training-content summary, copyright policy, or Article 55 documentation is published for GLM, and the 355B MoE's systemic-risk status is unconfirmed.