GLM-5.3 Launches With Frontier Coding and Autonomous Cyber Exploit Capabilities

Z-ai launches GLM-5.3 featuring emergent multi-stage cyber exploitation capabilities across real codebases. In real-world security testing across 269 projects, the model uncovered 2,436 vulnerabilities with an average lifespan of 26.6 years. Benchmark scores rose from 4.6 to 28.3 on Terminal-Bench 3.0 and from 46.2 to 66.9 on DeepSWE v1.1. On CyberGym, it reached 84.5%, passing Mythos 5 at 83.8% and GPT-5.6 Sol at 83.6%. The release eliminated zero-thinking mode completely, requiring low, high, or max reasoning effort on all API calls. Every major lab's turning automated exploitation loops loose on legacy infrastructure whether maintainers are ready for it.