Z.ai Unveils GLM-5.3: Top Open-Weights Coding, Cyber Skills
Zhipu/Z.ai presents GLM-5.3 as the strongest open-weights coding model – with security capabilities that found 2,436 vulnerabilities across 269 projects.
Illustration · AI-generated (AI IN LIFE)
At a glance
- GLM-5.3 was unveiled on August 14, 2026; Z.ai calls it the strongest open-weights coding model.
- Same base architecture as GLM-5.2 – gains come entirely from extended post-training.
- With security teams, the model found 2,436 vulnerabilities across 269 projects, some 40 years old.
- Per VentureBeat, GLM-5.3 has reportedly already found a serious vulnerability in Cursor.
- Weights are due as open source about two weeks later, after security reviews.
Chinese AI company Zhipu (Z.ai) has unveiled GLM-5.3, calling it the strongest open-weights coding model currently available. The release was covered by The Decoder, VentureBeat and Unite.AI, among others.
Technically, GLM-5.3 shares the same base architecture as GLM-5.2 – improvements come entirely from extended post-training, according to Z.ai. The company reports roughly a 50 percent performance gain over its predecessor, with particularly strong results on agentic coding tasks.
The security capabilities stand out: GLM-5.3 was specifically trained for vulnerability detection. Working with Chinese security teams, the model identified 2,436 vulnerabilities across 269 projects, the company says – including legacy flaws up to 40 years old. According to VentureBeat, the model has also reportedly found a serious vulnerability in the Cursor code editor.
Z.ai describes behavior that outgrew its training: the model reportedly began reasoning across multiple stages of exploitation, forming coherent plans for complete exploit chains. That is precisely why the company is tying the release to controls.
The model weights are scheduled for open-source release about two weeks after the announcement, pending security reviews. Until then, GLM-5.3 is available via the GLM Coding Plan and integrates with coding agents such as ZCode, Claude Code and OpenCode.
FAQ
Is GLM-5.3 freely available yet?
Not fully: it can be used via the GLM Coding Plan and integrations. The open model weights are scheduled to follow about two weeks after the announcement, once security reviews are complete.
What is special about its cyber capabilities?
The model was trained for vulnerability detection and developed the ability to plan multi-stage exploit chains – stronger than intended, per the company. That is why release is gated by controls.
Who is GLM-5.3 for?
Primarily developers: it targets agentic coding and integrates with tools like ZCode, Claude Code and OpenCode.
