Z.AI releases GLM-5.3, a 700B open-weight model with strong coding and cybersecurity benchmarks

Z.AI released GLM-5.3, a 700B MoE model targeting coding leadership. New articles confirm it achieves 84.5% on CyberGym, beating US frontier models, and found a 40-year-old vulnerability. Open weights expected in two weeks. Safety-first delayed release with 'trusted access' program. This is a major open-weight cybersecurity-capable model, part of the Chinese MoE surge.