Ethical Stack

GLM-5.3 family

Hosted model family for coding and long-horizon agents. GLM-5.3 and GLM-5.3-Flash are the latest verified family releases; the main 5.3 announcement said weights were planned after safety review, so current weight status should be rechecked per checkpoint.

Score calculation

(4) ÷ 1 = 4.0

1 of six dimensions rated. Unrated dimensions and BDS status are excluded.

Full scoring method ↗
Setup details

The release discloses model-development and human-in-the-loop details but not worker conditions. Performance/cyber capabilities and announced open weights do not establish downstream use or license scope.

Z.ai announced GLM-5.3 on August 14 and GLM-5.3-Flash on August 26, 2026. Its product pages list GLM-5.3 as latest flagship; checked 2026-10-23.

Energy

↑

GLM-5.3 release describes increased post-training compute and token efficiency but gives no training kWh, water, emissions, or measured inference energy. Token counts and speed do not establish environmental impact.

Reasoning & sources

Z.ai describes additional post-training compute and token efficiency but does not publish electricity, emissions, or water measurements. More efficient token use can coexist with larger total compute; comparing footprints requires training and serving measurements.

No model-specific energy or lifecycle figures found.

Training rights

↑

The release describes scaling RL on accumulated task environments and says human-in-the-loop work remains meaningful; it does not itemize pretraining/post-training data sources or rights for GLM-5.3.

Reasoning & sources

The release explains RL environments but does not enumerate pretraining or post-training data rights. Human-in-the-loop environment work is not evidence of data permission; request separate provenance for both data and interaction traces.

Need exact corpus provenance, licenses, and rights/consent policy.

Privacy

↑

The release offers hosted API/Coding Plan access and future local weights but does not define prompt training use, retention, human access, or data residency. Privacy depends on selected product and eventual self-hosting setup.

Reasoning & sources

The release offers hosted API and coding access but does not specify prompt retention, training use, human access, or residency. A privacy judgment would require the current service policy and product-specific terms; do not infer privacy from future local weights.

Need current endpoint-specific privacy and data-use terms.

Openness

↑

Z.ai described planned weight release for GLM-5.3 after safety evaluation; the checked release source did not independently establish final availability or exact license. Do not yet score as open-weight without the artifact/license.

Reasoning & sources

Z.ai announced planned weight release after safety evaluation, but the record does not verify completed publication or exact license. Until the checkpoint and its terms are publicly available, describe access as hosted rather than open-weight.

The company announced planned open weights, but the checked source did not verify final artifact/license availability.

Verify checkpoint weight publication and license at current official repository.

Worker treatment

↑

Z.ai says GLM-5.3's environment generation and verification still require meaningful human-in-the-loop work. It does not disclose worker count, pay, vendor chain, conditions, or protections.

Reasoning & sources

Z.ai says human-in-the-loop work remains meaningful in environment creation and verification, but provides no worker counts, pay, or vendor chain. Human contribution is documented, but worker treatment is unknown; request labor disclosures.

Human work is acknowledged; labor conditions remain undisclosed.

Government & military

↑

Z.ai publishes GLM-5.3 weights and local-serving instructions under its GLM-5.3 license. Independent deployment is technically possible subject to those terms; the checked sources do not identify a military customer or deployment.

Reasoning & sources

The official Z.ai model card includes model files and local Transformers/vLLM/SGLang serving instructions, but the license is a model-specific GLM-5.3 license rather than an unrestricted open license. Those facts make third-party deployment possible under its terms; they do not establish a military relationship.

Needs procurement and company-source review; no downstream-use register found.

BDS status

↑

We did not find Z.ai among the named targets in the two official BNC campaign lists. This is a limited list check, not BDS approval.

Reasoning & sources

Z.ai is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Running downloaded weights locally is a different decision from paying the developer or a cloud host; check the host as well.

This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.

All sources

Side by side