Zhipu AI Releases GLM-5.3 Model to Rival US AI Leaders
Zhipu AI released its GLM-5.3 AI model focusing on coding and cybersecurity, planning to open-source the model weights within two weeks to compete with OpenAI and Anthropic.
Beijing-based Zhipu AI released its flagship GLM-5.3 large language model on August 14, 2026, targeting improvements in coding and cybersecurity to compete with US-based leaders OpenAI and Anthropic. The model, an evolution of GLM-5.2, utilizes a mixture of experts architecture with 753 billion parameters and a 1 million token context window. It is currently available via API and the GLM Coding Plan, with model weights scheduled for release on Hugging Face within two weeks following safety assessments.
Zhipu AI claims GLM-5.3 outperformed OpenAI's GPT-5.6 Sol and Anthropic's Mythos 5 on the CyberGym benchmark, achieving an 84.5 percent success rate in identifying security flaws. The company reports the model has identified 2,436 vulnerabilities across 269 open-source projects. However, the model trailed Anthropic's flagship model on ExploitBench, which measures the ability to convert flaws into working attacks, scoring 54.4 percent compared to Mythos 5's 78 percent.
Alongside the release, Zhipu AI launched the Open Source Shield initiative to integrate code-auditing functions into its ZCode product. This move comes as other Chinese firms, including Alibaba Group and DeepSeek, attempt to narrow the performance gap with US frontier models. Despite the technical launch, Zhipu AI shares fell up to 9 percent in Hong Kong, a decline analysts attributed to profit-taking ahead of quarterly results.
