Zai.org releases GLM-5.3, an open-weight model with enhanced capabilities

Zai.org has released GLM-5.3, an open-weight model that demonstrates significant improvements over GLM-5.2. This new model excels in complex coding and long-horizon tasks, achieving a 50% improvement on the Z.ai Code Bench. Additionally, GLM-5.3 has attained state-of-the-art performance on public benchmarks such as Terminal Bench 3.0 and Agents' Last Exam.

Superior Coding Abilities

GLM-5.3 is the most capable open-weights model for coding, with a 50% improvement over GLM-5.2 on the Z.ai Code Bench. This advancement is attributed to its enhanced ability to handle complex coding tasks and long-horizon challenges.

Enhanced Cybersecurity Features

GLM-5.3 has demonstrated superior cyber capabilities, particularly in vulnerability discovery. It has achieved state-of-the-art performance on CyberGym for vulnerability discovery, with significant gains further up the exploitation chain.

Adjustable Reasoning Effort

GLM-5.3 supports controlling the thinking budget through the reasoning_effort parameter, which accepts three levels: low, high, and max. This feature allows users to adjust the model's performance based on the complexity of the task at hand.

Zai.org's GLM-5.3 release marks a significant step forward in open-weight models, offering enhanced capabilities in coding and cybersecurity. With its superior performance on various benchmarks, GLM-5.3 is poised to make a notable impact in the AI community.