[Zhipu Releases GLM-5.3 Model, Excelling in Complex Coding and Long-Term Tasks] BlockBeats News, August 14, Zhipu (02513.HK) announced the release of GLM-5.3 today. It utilizes the same foundational model as GLM-5.2—all performance improvements stem from post-training optimizations. Compared to GLM-5.2, it performs better in complex coding and long-term tasks. GLM-5.3 boasts stronger coding capabilities and is currently the most powerful open-source weight model. In our internal Z.ai code benchmarking, its performance improved by 50% over GLM-5.2. Additionally, during large-scale deployment after training, its network capability development exceeded our expectations. On the CyberGym platform, GLM-5.3 leads in vulnerability detection, with the most significant improvements observed in the latter stages of exploitation chains—its performance in exploitation benchmarks is more than double that of GLM-5.2. Once safety assessments and reinforcement work are completed, we will release the weights two weeks after the announcement. [Original Link]