Model
GLM
Publisher
Zhipu AI (Z.ai) (China)
Licence
MIT License (GLM-4.5 / GLM-4.6 / GLM-4-0414 series; verify per checkpoint)
Context
200K / up to 128k (per checkpoint) / 32K native, up to 128K extended
GLM is Zhipu AI's open-weight family - Zhipu AI is the Beijing company (spun out of Tsinghua's THUDM lab) that ships internationally under the "Z.ai" brand.
Do you really own it?
Partial
none·limited·partial·substantial·full
Analytical input: AOI C · 65.2/100
Floor-weighted, not averaged. The weakest factor caps the level, because ownership is a conjunction. Transparency is weak, which holds the whole verdict down regardless of the rest.
You substantially - but not uniformly - own the GLM line, and the split is per-variant. The current line is genuine MIT: GLM-5.2 ships a real checked-in MIT LICENSE ('Copyright (c) 2026 Zhipu AI') and the GLM-4.5/4.6 and GLM-4-0414 siblings declare the same MIT via HF metadata/README, so you can run, fine-tune, distill, redistribute and commercialise them with no gate. The caveat is the legacy glm-4-9b - NOT plain MIT but a custom 'glm-4' licence requiring commercial-use registration, 'Built with glm-4' attribution and a name prefix, governed by PRC law with disputes to Beijing's Haidian District court - so verify the LICENSE per checkpoint. Two things hold it at partial: training is opaque (closed data/code, China-aligned alignment you cannot inspect) and no safety battery is benchmarked, so supply your own guards. Mind the jurisdiction split too - the weights you self-host are yours, but the hosted Z.ai service is Singapore-governed (SIAC arbitration), a different legal regime from the PRC glm-4-9b weights licence; do not conflate the two.
1
Use and modify freelyCan you run, modify and adapt it with no gate and no field-of-use trap?
StrongPer-variant, but clean on the current line: GLM-5.2 ships a real checked-in MIT LICENSE ('Copyright (c) 2026 Zhipu AI', no custom clauses) and the GLM-4.5/4.6 and GLM-4-0414 siblings declare the same MIT via HF metadata/README, so you can run, fine-tune, distill, redistribute and commercialise the 9B-to-355B ladder with no gate. The exception is the legacy glm-4-9b, which is NOT plain MIT - a custom 'glm-4' licence with commercial-use registration, 'Built with glm-4' attribution and a name-prefix requirement under PRC law - so verify the LICENSE per checkpoint.
How this scores (AOI sub-dimensions)
Openness3/5how much is released - weights, data, code, licence - and how freelyOpen-weights tier: weights are downloadable under a genuine MIT license with model cards and technical reports, but the training data and training code are not released (only inference code) and evaluation is only partially reproducible.
Legal3/5how permissive and clean the licence is for real commercial useThe OSI-approved MIT license on the GLM-4.5/4.6 and 0414 series is a real plus - cleaner than a restricted community license.
2
TransparencyDo you know what it is: weights, training, behaviour, and legible terms?
WeakYou hold the weights but cannot see what they are: training data and training code are closed, the models carry China-aligned alignment you cannot inspect, 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 modelVerified zai-org on Hugging Face plus official ModelScope publication, safetensors with checksums, and no canonical-org malicious incident (checklist 5/8).
Governance3/5how accountable and well-documented the publisher isActive, accountable publisher (Zhipu AI / Z.ai) with a rapid release cadence and a verified presence on Hugging Face and ModelScope, but no documented vulnerability-disclosure or deprecation policy and no EU Code of Practice signature.
3
Doesn't fail youIs it reliable and good enough for the job?
ModerateStrong coding/agentic capability (GLM-4.6 is among the more capable open models on third-party evaluation), but safety tuning is lighter than Western frontier labs with no companion guard model, and the 355B MoE carries an unresolved EU systemic-risk question.
How this scores (AOI sub-dimensions)
Performance4/5how capable it is relative to its classStrong general capability and especially strong coding/agentic performance: third-party evaluation places GLM-4.6 among the more capable open models for real-world coding and tool use.
Operational4/5how practical it is to run, serve and maintain in productionStrong ecosystem support: 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, but tuning is lighter than Western frontier labs, there is no companion guard model, the hosted API applies Chinese content moderation, and the models carry China-aligned alignment - behavioural factors to account for.
4
Doesn't extract your dataDoes running it keep your knowledge and data yours?
StrongSelf-hosted, the MIT-licensed weights run on your own infrastructure and nothing phones home - your data and derived knowledge stay yours. The hosted Z.ai API is the only path that exposes inputs, and it is a Singapore regime (not PRC): API content is 'not saved on our servers' and not used to train unless you agree, 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.
How the AOI score is computed
The seven dimensions above, each scored 0 to 5, weighted and summed to the 0 to 100 headline. The score is the analytical input behind the ownership verdict, not the verdict itself.
DimensionScoreWeightPoints
Openness3/50.1810.8
Provenance3/50.169.6
Legal3/50.169.6
Safety3/50.169.6
Performance4/50.1411.2
Operational4/50.129.6
Governance3/50.084.8
HeadlineC · 65.2/100
Grade ceiling: a hard flag (🚩 EU AI Act - the 355B GLM-4.5/4.6 MoE may approach the 1e25-FLOPs systemic-risk threshold (unconfirmed) and no Article 55 / training-content documentation is published) caps the grade below the raw band. See the classification matrices.
Sources
Every rating traces to a primary document. Read means the text was verified; unverified means it is known to exist but has not yet been read.
DocumentWhat it grounds
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).
Licenceread2026-07-25
The GLM-4.5, GLM-4.6 and GLM-4-32B-0414 checkpoints declare the OSI-approved MIT license permitting commercial use - but via HF metadata ("License: mit") and README only; there is NO in-repo LICENSE file (raw/blob LICENSE returns 404).
Licenceread2026-07-25
GLM-5.2 ships a REAL in-repo MIT LICENSE file, read verbatim: standard MIT text with no custom clauses, "Copyright (c) 2026 Zhipu AI." This resolves the metadata-only gap on the 4.5/4.6/0414 line by confirming the family MIT is a genuine, checked-in MIT licence on at least one current sibling.
Licenceread2026-07-25
Licensing is per-variant and the legacy glm-4-9b is NOT plain MIT.
Documentationunverified2026-07-25
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 serviceread2026-07-25
Hosted Z.ai Terms of Service, read verbatim: governed by Singapore law with SIAC arbitration (seat Singapore).
Privacy Policyread2026-07-25
Hosted Z.ai Privacy Policy, read verbatim: data is processed in Singapore; API content is "not saved on our servers"; consumer content is used "when we train and improve our models." The Singapore processing location is distinct from the PRC jurisdiction of the glm-4-9b weights licence.
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.