8.3
GLM-5.3 Review — Post-Training Scaling Puts Open Weights Within a Hair of Mythos 5
GLM-5.3 is Z.ai's new frontier flagship built on the GLM-5.2 base with post-training only. Terminal-Bench 3.0 jumps 4.6→28.3, DeepSWE v1.1 46.2→66.9, Agents' Last Exam 23.8→28.5. It scores 84.5% on CyberGym (vs Mythos 5's 83.8%) and 54.4% on ExploitBench (up from 24.4%), and its security sweep found 2,436 vulnerabilities across 269 open-source projects. Weights ship in two weeks; API 'coming soon'; Coding Plan subscribers get it today. Review covers benchmarks, the Z.ai Code Bench private eval, the security disclosure ledger, and the HN debate on whether open cyber-capable models change the calculus.