Model
GLM-4-9B-Chat
Publisher
Zhipu AI (Z.ai / THUDM) (China)
Licence
glm-4 Model License
You have only limited ownership of the original GLM-4-9B-Chat: the custom glm-4 licence is revocable, gates commercial use behind registration, restricts fields of use, and forces a 'glm-4' name prefix and 'Built with glm-4' attribution on every derivative - so use-and-modify is weak and your continued-use control is contingent. Combined with dated capability, ownership lands at limited, the low end of the GLM family and well below the MIT GLM line (the `glm` entry, substantial). Use this checkpoint only if you specifically need it: complete the commercial registration, honour the naming and field-of-use terms, and account for the revocable grant - otherwise adopt the MIT GLM models instead.
Do you really own it?
Limited
none·limited·partial·substantial·full
Analytical input: AOI D · 52.8/100
The four ownership factors
Floor-weighted, not averaged. Use & modify is weak, and the weakest factor sets the ceiling, so the verdict stays limited however strong the rest.
1
Use and modify freelyCan you run, modify and adapt it with no gate and no field-of-use trap?
WeakThe glm-4 licence is non-OSI and REVOCABLE, gates commercial use behind a registration form, forbids military/illegal/national-security use, and forces every derivative to display 'Built with glm-4' and to prefix its name with 'glm-4'. A revocable, registration-gated grant with mandatory naming is the weakest use-and-modify position in the GLM family - weak.
How this scores (AOI sub-dimensions)
Openness2/5how much is released - weights, data, code, licence - and how freelyGated-open: the weights are downloadable and free for academic research, but commercial use is registration-gated and the grant is revocable, with training data and code closed.
Legal2/5how permissive and clean the licence is for real commercial useThe worst legal position in the GLM family: a non-OSI licence that is REVOCABLE, gates commercial use behind registration, restricts fields of use, mandates "Built with glm-4" and a name prefix, and is governed by PRC law (Haidian court) - with no EU copyright policy or training-content summary and no open-source exemption.
2
TransparencyDo you know what it is: weights, training, behaviour, and legible terms?
ModerateWeights are inspectable and there is a model card, but training data and code are closed and the model carries China-aligned alignment you cannot inspect - open weights but closed process, so moderate.
How this scores (AOI sub-dimensions)
Provenance3/5how well we can trace and verify what went into the modelVerified THUDM org on Hugging Face, safetensors with checksums, clear canonical source, no malicious-checkpoint incident (checklist ~4/8).
Governance3/5how accountable and well-documented the publisher isActive, named publisher (Zhipu AI / THUDM) with a verified org and a licence contact (license@zhipuai.cn), meeting the score-3 anchor.
3
ReliabilityIs it reliable and good enough for the job?
ModeratePerformance 2 (dated, superseded), operational 4, safety 3: runnable and broadly supported, but the below-average current capability holds reliability to moderate.
How this scores (AOI sub-dimensions)
Operational4/5how practical it is to run, serve and maintain in productionA small 9B dense model with broad serving (vLLM, transformers, llama.cpp, Ollama) and extensive community quants, runnable on a single GPU.
Safety3/5whether misuse risks are evaluated and guardrails are providedA safety-tuned chat model with documented behaviour, meeting the score-3 anchor, but with China-aligned alignment, no first-party guard model, and older/lighter tuning than current frontier practice - deployers must add their own guardrails.
4
Doesn't extract your dataDoes running it keep your knowledge and data yours?
ModerateSelf-hosting keeps your processing local, but the licence grant is revocable and commercial use hinges on a registration the publisher controls - so your control over continued use is contingent, not assured. Moderate, not strong.
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
Openness2/50.187.2
Provenance3/50.169.6
Legal2/50.166.4
Safety3/50.169.6
Performance2/50.145.6
Operational4/50.129.6
Governance3/50.084.8
HeadlineD · 52.8/100
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
Licenceread2026-08-03
THUDM/glm-4-9b LICENSE, read: [Sec 2] "a non-exclusive, worldwide, non-transferable, non-sublicensable, revocable, royalty-free copyright license"; free for academic research; "Commercial users must complete registration ...
Model cardread2026-08-03
GLM-4-9B-Chat model card on the verified THUDM HF org: a 9B dense chat model with a 128K context (and a 1M-context GLM-4-9B-Chat-1M variant), safetensors; broad serving (vLLM, transformers, llama.cpp, Ollama) and extensive community quants; a mid-2024 release.
Third-party analysisunverified2026-08-03
GLM-4-9B-Chat was a capable small model at its mid-2024 release but is dated by 2026 and superseded within its own family by GLM-4-9B-0414 and the 4.5/4.6 line.
Third-party analysisunverified2026-08-03
GLM chat models carry China-aligned alignment on politically sensitive topics and ship no companion guard model; the original 9B's safety tuning is older/lighter than current practice.
Third-party analysisunverified2026-08-03
The glm-4 licence is revocable and gates commercial use behind registration with field-of-use and naming restrictions (non-FOSS), so no open-source exemption applies; no EU copyright policy or training-content summary is published, and the licence is governed by PRC law.