Z.ai's GLM-5.3: Why Full AI Model Release is Delayed

Z.ai has launched GLM-5.3, a new version of its flagship AI model, but is holding back the full model weights. The company states GLM-5.3 offers significant improvements in coding and cybersecurity. These gains come from larger-scale post-training in diverse and complex working environments, without changing the underlying architecture of GLM-5.2. The model is now available through its API and Coding Plan. However, the full model weights will be released in about two weeks, after additional safety testing. Z.ai reports GLM-5.3 scored 28.3% on Terminal-Bench 3.0, a sharp rise from 4.6% for GLM-5.2. In cybersecurity, its score on CyberGym reached 84.5%, up from 77.2% for the previous version. The model also more than doubled its score on ExploitBench to 54.4%. This development shows how AI models are rapidly advancing in complex problem-solving, which could impact various technical fields.