Z.ai Releases GLM-5.3 Weights — With License Caveats

Z.ai has announced the release of its GLM-5.3 model in open-weight form: the weights can be downloaded, run locally and tuned to your own tasks. The files are published on Hugging Face, with an accompanying technical breakdown in the developers' blog.

The announcement describes GLM-5.3 as the most capable model in the lineup for two scenarios: agentic coding (that is, working as an autonomous developer assistant) and cyberdefense. No detailed benchmarks are given in the announcement itself — for the specifics, the authors point readers to the technical blog.

Running locally and the license

An open release of the weights is immediately picked up by the community. The Unsloth AI team congratulated Z.ai on yet another release and said it was preparing quantized builds in GGUF format — convenient to run on home hardware for those with enough computing resources.

The discussion, however, drew attention to an important nuance: GLM-5.3 moves away from the previously permissive MIT license toward a “semi-commercial” distribution model — with restrictions similar to those already used by Kimi, MiniMax, Qwen and a number of others. In other words, “open weights” here is not the same as full freedom of commercial use, and the terms are worth reading carefully before deployment.

This shift reflects a broader trend: Chinese labs are happy to release the weights of their models, but increasingly attach licenses that specifically restrict commercial use, reserving control over monetization for themselves.

Source: Hacker News