Z.ai's GLM-5.3: a 743B coding model with emergent cybersecurity skills

Z.ai released GLM-5.3, a 743B-parameter coding model built on the same base as GLM-5.2, with every gain coming from extended post-training. The company reports a 50% lift on its internal coding benchmark and 84.5% on CyberGym, a vulnerability-reasoning test, plus 2,436 flaws found across 269 projects. Weights are due roughly two weeks after launch, pending safety hardening.
3 sources
- 01The Decoder — Zhipu AI releases GLM-5.3, claims it's the strongest open-weights coding modelthe-decoder.com
- 02InfoWorld — Zhipu says new coding AI developed advanced cyber skills faster than expectedinfoworld.com
- 03MarkTechPost — Z.ai Ships GLM-5.3 Without Retraining the Base Model: Better at Complex Coding and Long-Horizon Tasksmarktechpost.com