GLM-5.3 is an open-weights large language model released by Z.ai, built on the same base model as GLM-5.2 with all of its improvements coming from scaled post-training rather than a new pre-train. That post-training stack combines IndexShare for efficient long-context processing, SAO for reinforcement learning on long-horizon tasks, and slime for large-scale asynchronous training, all applied across an expanding set of long-horizon task environments. The result is a model aimed squarely at complex software engineering and multi-step agentic workflows, where it can keep large contexts coherent while carrying out extended tool-using trajectories.
In qualitative terms, GLM-5.3 is positioned as the most capable open-weights model for coding, with a roughly fifty percent gain over GLM-5.2 on the in-house Z.ai Code Bench and open-source state-of-the-art results on Terminal Bench 3.0 and Agents' Last Exam. The released benchmark table also shows competitive scores against frontier closed models such as Opus 4.8 and GPT-5.6 Sol on Terminal Bench 2.1, DeepSWE v1.1, SWE-Marathon v1.1, and PostTrainBench. A notable side effect of scaling post-training is emergent cyber capability: GLM-5.3 reaches state-of-the-art performance on CyberGym for vulnerability discovery and more than doubles GLM-5.2 on exploitation-chain benchmarks, making it well suited for security research and vulnerability analysis workloads as well as general agentic coding.
