Z.ai releases GLM-5.3's open weights by August 31st
Why I called it
Z.ai says it is holding the back roughly two weeks for safety hardening, after reporting GLM-5.3 found 2,436 real vulnerabilities across 269 open-source projects and scored above its own reading of two closed frontier competitors on CyberGym. That is a short, self-imposed window from a lab that has shipped every prior GLM generation openly, and whether two weeks means two weeks is a data point on how a Chinese lab handles a model whose offensive capability outran its own testing.
The call, in full. Z.ai publishes downloadable GLM-5.3 weights on Hugging Face or ModelScope by August 31st 2026.
Scoring criterion. RESOLVES CORRECT if a GLM-5.3 weights repository published by Z.ai is publicly downloadable on huggingface.co or modelscope.cn on or before 2026-08-31 23:59 UTC. RESOLVES WRONG otherwise.
The criterion is the machine-checkable version: a prediction that cannot be settled by a third party against a public source fails the build before it reaches this page.