GLM-5.3: 50% Better Coding, Challenges Frontier AI

Z.ai's new GLM-5.3 model shows a 50% improvement on its internal coding benchmark without increasing its size. This is achieved through expanded post-training on more environments and diverse tasks. What's interesting is that GLM-5.3 scored 91.25% on the KingBench 3 coding benchmark, outperforming its predecessor, GLM-5.2, which scored 75%. It also surpassed other models like Fable 5 and Opus 4.8 on this specific test. While GLM-5.3 is competitive across several broader coding benchmarks, it doesn't dominate every single one. For example, it nearly matches other top models on Terminal-Bench 2.1. On the harder Terminal-Bench 3.0, GLM-5.3 shows a significant jump from its previous version, scoring 28.3 compared to GLM-5.2's 4.6. This indicates that Z.ai is making substantial progress in AI coding capabilities, which could impact various tech applications.