Zhipu AI's GLM 5.3 Boosts Coding and Cybersecurity Skills via Post-Training Alone

Chinese AI firm Zhipu AI released GLM 5.3 on August 14, 2026, claiming significant performance gains without changing the underlying model architecture or size. The model retains the same approximately 744-billion-parameter Mixture-of-Experts design as its predecessor GLM 5.2, with all improvements attributed solely to scaled post-training methods. Zhipu reports a 50% improvement over GLM 5.2 on its internal coding benchmark and top open-source scores on Terminal-Bench 3.0, while also achieving state-of-the-art results on the CyberGym vulnerability-discovery benchmark. The gains were enabled by more diverse training environments, a 2.3x improvement in RL training throughput, and better token efficiency — producing stronger results with fewer output tokens. Model weights are expected to be released publicly within two weeks, and all benchmark figures currently reflect vendor-reported data pending independent verification.