GLM-5.3

GLM-5.3 is the flagship open language model from Z.ai. The developer presented it on 2026-08-18, and the weights are published on Hugging Face. It is built on the same base model as GLM-5.2: according to the model card, all gains come from additional training afterwards. The model has about 753 billion parameters. It works with text only, and its context window is one million tokens, with up to 128 thousand tokens in a reply.

The model is aimed at programming, long agent tasks and security work. The authors say it improves on GLM-5.2 by 50% on their own coding test and show large gains on terminal and agent tests; these are the developer's own results, not independent checks. Reasoning is always on and cannot be switched off; its depth can be set to low, high or maximum, and maximum is the default. The card lists English and Chinese. It runs with Transformers, vLLM, SGLang, TokenSpeed, KTransformers and Unsloth, and Ascend NPUs are supported.

The weights are public without any acceptance step. The GLM-5.3 License lets you use, change, fine-tune, distribute and sell the model for free; you only need to keep the license notice and follow the law. One condition is aimed at very large API providers: if you run a business that gives third parties access to model inference or fine-tuning, for example through an API, and your revenue together with affiliates is above 10 billion US dollars over 12 months, you must pass the developer's security review before commercial use. End-user products that only embed the model in some feature, and simple relaying of requests to models hosted by others, are not counted. Because of this condition we mark commercial use as limited. We found no territorial restrictions in the license text; read the license before use.

This is not the same model as GLM-5.3-Flash: that one is smaller (320 billion parameters), understands images as well as text and is released under the MIT license. The card of this model does not state hardware requirements; with this size it is a server-level model.

Date of check: 2026-10-31.