Z.ai releases open-weight GLM-5.3, topping agentic benchmarks

GLM-5.3 hit 84.5% on the CyberGym vulnerability benchmark and 60 on the Artificial Analysis Intelligence Index, tying top open-weight models. It beats GPT-5.6 and Claude Fable 5 on agentic benchmarks with a 50% gain over GLM-5.2 on Code Bench, achieved via post-training on the GLM-5.2 base.