Zhipu AI has released GLM-5.3, its latest coding model, which it claims is the strongest open-weights coding model to date. The model is based on the same foundation as its predecessor, GLM-5.2, but incorporates enhanced post-training to achieve significant improvements. According to Zhipu, GLM-5.3 shows the biggest gains in agent-based tasks, making it a strong contender in the coding model space. The company emphasized that the model's enhancements are focused on improving performance in complex, multi-stage tasks. The release comes as part of Zhipu's ongoing efforts to strengthen its position in the AI development landscape. Zhipu AI also highlighted the model's potential in cybersecurity, where it has been trained to identify software vulnerabilities. The company stated that GLM-5.3 is now available through the GLM Coding Plan and can be used with coding agents like ZCode, Claude Code, or OpenCode. The model weights are expected to go open source in two weeks after completing security reviews. Source: thedecoder
Zhipu trained GLM-5.3 with data and environments designed to detect software vulnerabilities. The model is capable of reasoning across multiple stages of exploitation, forming coherent plans for complete exploitation chains. Working with security teams in China, Zhipu identified 2,436 vulnerabilities across 269 projects, some of which are up to 40 years old. These flaws are documented in a public registry. The company's efforts in cybersecurity highlight the practical applications of GLM-5.3 beyond traditional coding tasks. Source: thedecoder
Zhipu AI claims GLM-5.3 is the strongest open-weights coding model, with the biggest improvements in agent-based tasks. The model's enhancements are attributed to extended post-training alone, without changes to the base architecture. Zhipu also announced that the model weights will be open-sourced in two weeks after security reviews are completed. Source: thedecoder
