Alpha
Contact
Assess · GLM

Can you own it?

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

This page is a projection of the one entry record, the Use & modify and Transparency factors that Assess 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

Intended & out-of-scope use

GLM is Zhipu AI's open-weight family. Zhipu is the Beijing company, spun out of Tsinghua's THUDM lab, that ships internationally under the Z.ai brand. The largest model tracked here, GLM-4.6, is a 355B-total / 32B-active Mixture-of-Experts model with a 200K context window, tuned for coding and agentic use. The family also includes GLM-4.5 (355B/32B), the lighter GLM-4.5-Air (106B/12B), and the dense GLM-4-32B-0414 and GLM-4-9B-0414.

Intended use is general-purpose assistant work, with a clear tilt toward coding, tool use and agentic workloads, where GLM-4.6 is strongest. Deploy the instruct checkpoints behind your own guardrails. Out of scope without additional controls: EU high-stakes use of the 355B MoE while the systemic-risk question is unresolved (see EU AI Act below), and any deployment that assumes MIT terms without checking the per-checkpoint LICENSE, because the original glm-4-9b used a custom non-OSI license.

Known limitations, bias & failure modes

The hosted Z.ai API applies Chinese content moderation on politically sensitive topics, and the open weights carry China-aligned alignment. Account for this behaviourally instead of assuming Western-lab defaults.

Safety tuning is lighter than at the large Western labs. Instruct alignment withstands casual jailbreaks, and no companion guard or classifier model ships.

Capability leadership is strongest in coding and agentic tasks and is not uniform across domains.

The training corpus is not released, so it cannot be inspected for known problem sources and the model is an opaque artifact.

Openness tier & components

GLM is open_weights. Open: downloadable weights, model cards, released inference code and technical reports. Closed: the training data and the training pipeline. Partial: evaluation is only partially reproducible.

The GLM-4.5/4.6 and 0414 series ship under the OSI-approved MIT license (ev-license-mit), an advantage over the restrictive community licenses some China peers use. No independent openness classification backs this here, so Dimension 1 rests on publisher evidence and scores 3.

License terms & permitted use

The GLM-4.5, GLM-4.6 and GLM-4-0414 series are MIT: OSI-approved, with no field-of-use restriction and unconditional commercial use. Derivatives can be used, modified, redistributed and commercialised subject only to the MIT attribution terms.

Verify the LICENSE per checkpoint. Licensing is per variant: the newer families are MIT, but the original glm-4-9b shipped under a custom, non-OSI "glm-4" license (ev-license-split). Read the LICENSE file on the exact model before use. This per-variant picture feeds the legal score (3).

The tracked cluster is uniformly MIT and ungated. The 0414 repos ship an in-repo MIT LICENSE file, while GLM-4.5 and GLM-4.6 rely on Hugging Face metadata and the README (ev-license-mit, ev-0414). Outside this entry, Zhipu's newer GLM-5.3 and GLM-5.3-BF16 checkpoints carry a custom licence that requires a Z.AI security review from Model-as-a-Service operators with more than 10 billion USD of group revenue over 12 months (ev-license-glm53). GLM-5, 5.1, 5.2 and 5.3-Flash are plain MIT.

Supply-chain & provenance

Weights are distributed from the verified zai-org org on Hugging Face in safetensors with per-file checksums, and dual-published on Zhipu's ModelScope (ev-modelscope). Two supply-chain surfaces need care. The dual-hub distribution requires checksums to match across Hugging Face and ModelScope. The legacy THUDM org still holds older GLM/ChatGLM checkpoints, while the current canonical org is zai-org.

The checkpoint trust checklist scores 5/8. Provenance scores 3: a scanned checkpoint or a documented mirror and quantization policy would be needed for 4, and cryptographic signing for 5. There is no weight signing or SLSA attestation. No incidents are on record for the canonical org. Pin the exact revision and verify checksums on download.

EU AI Act posture

GLM is a GPAI model. On licence grounds the MIT-licensed sizes are free and open-source releases with public parameters and usage information, so they plausibly qualify for the Article 53 open-source exemption from the Annex XI/XII technical-documentation duties. Two caveats keep this partial.

First, systemic risk is unresolved. The 355B GLM-4.5/4.6 MoE may approach the 10²⁵-FLOPs threshold, but this is not publicly confirmed. If it crosses, the exemption is void for that model and the full Article 55 package (not published) would be owed. The smaller dense 9B and 32B are well under.

Second, surviving obligations are unmet. A copyright policy and a public training-content summary survive the exemption, and neither is published. The training corpus is not released, so a downstream provider cannot assemble the training-content summary from upstream artifacts.

A China-based provider is unlikely to furnish an EU AI Office documentation package. An EU deployer receives an MIT licence and a usable model card, but must assemble the compliance material and resolve or avoid the systemic-risk question for the 355B MoE.

Benchmarks & evaluation

GLM-4.6 evaluates as strong on real-world coding, tool use and agentic tasks and competitive in general capability. Its coding and agentic strength is the main differentiator. The picture is aggregated from third-party evaluation (ev-perf-coding). Published scores are third-party and in-class, not a claim of uniform category leadership. GLM's own evaluation is only partially reproducible, because the harness and data are not fully open.

How this scores

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

1

Use and modify freelyCan you run, modify and adapt it with no gate and no field-of-use trap?

Strong

Licensing is per variant, and the current line is clean. GLM-4.6 ships a checked-in MIT LICENSE ('Copyright (c) 2026 Zhipu AI', no custom clauses), and the GLM-4.5 and GLM-4-0414 siblings declare the same MIT via HF metadata and README. The 9B-to-355B ladder can be run, fine-tuned, distilled, redistributed and commercialised with no gate. The exception is the legacy glm-4-9b, which is not plain MIT: its custom 'glm-4' licence has commercial-use registration, 'Built with glm-4' attribution and a name-prefix requirement under PRC law. Verify the LICENSE per checkpoint.

How this scores (AOI sub-dimensions)
Openness3/5how much is released - weights, data, code, licence - and how freelyThe tier is open-weights: weights are downloadable under an OSI-approved MIT licence, with model cards and technical reports.
Legal3/5how permissive and clean the licence is for real commercial useThe OSI-approved MIT licence on the GLM-4.5/4.6 and 0414 series is a clear advantage over a restricted community licence.
2

TransparencyDo you know what it is: weights, training, behaviour, and legible terms?

Weak

The weights are held but cannot be explained: training data and training code are closed, the models carry China-aligned alignment that cannot be inspected, and the hosted Z.ai API applies Chinese content moderation.

How this scores (AOI sub-dimensions)
Provenance3/5how well we can trace and verify what went into the modelThe zai-org org on Hugging Face is verified, with official ModelScope publication, safetensors with checksums and no malicious incident on the canonical org (checklist 5/8).
Governance3/5how accountable and well-documented the publisher isZhipu AI / Z.ai is an active, accountable publisher with a rapid release cadence and a verified presence on Hugging Face and ModelScope.
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.