Zhihui AI has released GLM-5.3, an open-weights model that builds on the GLM-5.2 base architecture through post-training improvements. The update significantly enhances performance in complex coding and long-horizon tasks, establishing new state-of-the-art results on several public benchmarks.
- Achieves a 50% improvement over GLM-5.2 on the Z.ai Code Bench and sets open-source SOTA on Terminal Bench 3.0 and Agents' Last Exam.
- Demonstrates emergent cyber capabilities, reaching state of the art on CyberGym for vulnerability discovery and more than doubling GLM-5.2 on exploitation benchmarks.
- Supports deployment via frameworks including SGLang, vLLM, TokenSpeed, Transformers, KTransformers, and Unsloth, with specific support for Ascend NPU platforms.
- Introduces a reasoning_effort parameter allowing users to control thinking budget with low, high, or max levels, defaulting to max.
The release provides developers with a highly capable tool for agentic engineering and complex software tasks, offering flexible deployment options and explicit controls over model reasoning depth.