AI BriefingZ.ai Releases GLM-5.3 and Delays Security Review
Key point
Z.ai released GLM-5.3 with enhanced cybersecurity capabilities but delayed the open-weight release for two weeks for safety review.
Details
Z.ai's latest model GLM-5.3 significantly improved performance through expanded post-training instead of new pre-training.
Key benchmark results are as follows:
- Terminal-Bench 3.0: 4.6 → 28.3
- DeepSWE v1.1: 46.2 → 66.9
- CyberGym: 84.5% (comparable to Claude Mythos 5 and GPT-5.6 Sol)
Cybersecurity performance is particularly notable. During evaluation, the model discovered 2,436 vulnerabilities (1,097 of which were critical or high severity) across 269 open-source projects, including Linux, WebKit, and FreeBSD.
As the model demonstrated unintended multi-stage attack planning and execution capabilities, Z.ai decided to delay the open-weight release for approximately two weeks for safety assessment. However, these figures are based on Z.ai's internal reports and require independent verification.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.