Zhipu AI has launched GLM-5.3, an upgraded open-weights AI model promising significant improvements in coding performance and vulnerability detection, with potential to challenge leading systems like Anthropic’s Fable 5 and OpenAI’s GPT-5.6.
Zhipu AI has unveiled GLM-5.3, a new version of its open-weights model that it says delivers a substantial advance in coding and agentic work. The company describes the upgrade as a post-training overhaul rather than a new base model, with longer runs, more task environments and a wider range of scenarios shaping the latest release. In company testing, GLM-5.3 is said to have improved coding performance by 50 per cent, while Z.ai says it now tops open-weight systems on Terminal Bench 3.0 and Agents’ Last Exam.
The model is already available through Z.ai’s API and its GLM Coding Plan, where it can be used with coding agents including ZCode, Claude Code and OpenCode. Zhipu AI has also suggested its coding and agent capabilities are approaching Anthropic’s Fable 5, although that comparison rests on the company’s own benchmarking. Axios reported that GLM-5.3 scored 84.5 per cent on the CyberGym benchmark, ahead of Anthropic’s Fable 5 and OpenAI’s GPT-5.6 Sol, though those results have not been independently verified.
Security is a major part of the pitch. Zhipu AI says the model was trained with vulnerability-research data and specialised environments so it can reason through multi-step exploitation rather than merely spotting flaws. According to the company, work with Chinese security teams has already helped identify 2,436 vulnerabilities across 269 projects, including defects in older software, and Axios said more than 1,000 were classed as critical. Zhipu AI plans to release the weights about two weeks after launch, after further safety evaluation and hardening, a delay that reflects the tension between wider access and the model’s offensive-security potential. That matters because GLM-5.2, the previous version, had already been reported as competitive in vulnerability detection, and Tom’s Hardware recently noted that Z.ai has built out domestic compute capacity with a 1-gigawatt data centre powered entirely by Chinese chips, underlining the scale of infrastructure behind the GLM line.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
