Zhipu AI releases GLM-5.3, claims top open-weights coding performance

Zhipu AI has released GLM-5.3, a coding-focused model that shares the same base as GLM-5.2. According to Zhipu, the gains come from extended post-training alone, including a claimed 50 percent improvement over its predecessor. The company says GLM-5.3 is the most powerful open-weights coding model, with the largest jumps in agent-based tasks.

The model was trained with data and environments designed to find software vulnerabilities. Zhipu says GLM-5.3 helped security teams in China identify 2,436 vulnerabilities across 269 projects, including some in projects up to 40 years old. The Decoder notes that cybersecurity is an area where top Chinese models such as Kimi and Qwen still trail US frontier models.

GLM-5.3 is available now through the GLM Coding Plan. The model works with coding agents including ZCode, Claude Code, and OpenCode. Its weights are scheduled to go open source in two weeks, after security reviews are completed.

GLM-5.3 is positioned as a post-training-driven upgrade rather than a new-base-model release. For developers, the most relevant claims center on agent-based coding performance and compatibility with existing coding-agent workflows. For security teams, the vulnerability-finding results are notable, but the claims come from Zhipu and should be evaluated against independent testing once the weights are released.