Yesterday we covered GLM-5.3's launch, and now Zhipu AI has published the model's weights on Hugging Face, making the frontier system downloadable and usable outside their API.
Zhipu AI confirmed the open-weight release: the model files are hosted under a GLM-5.3 License that allows download, local deployment, customization and commercial use, while requiring a security review for organizations above a revenue threshold. The release follows an API-only launch roughly two weeks earlier.
This changes who can test and use a reported frontier model. External researchers and companies can now run the exact system Zhipu described, reproduce coding and security benchmarks, and build custom versions instead of relying only on API queries.
Think of the weights as the machine's blueprints. With them, teams can run the model themselves, change its behavior, and integrate it into products. Running a frontier model still needs heavy GPU infrastructure and engineering, so 'local' usually means data-center or cloud setups, not a laptop.
What changes now is concrete: independent benchmarking should start quickly, developers and startups can build customized agentic coding and cybersecurity tools, and enterprises can deploy the model behind their own controls rather than routing every request through Zhipu. At the same time, the license's security-review and the compute cost will steer early adoption toward well-resourced teams.
The real test is whether independent teams confirm Zhipu's performance claims and whether companies turn those results into competitive products. We'll be watching benchmarks and early commercial integrations in the coming months.