BlockBeats News, August 14th, Zhifu (02513.HK) announced today the release of GLM-5.3. It uses the same base model as GLM-5.2— all performance improvements come from post-training optimization.

Compared to GLM-5.2, it performs better in complex encoding and long-task scenarios. GLM-5.3 has a stronger encoding capability and is currently the most powerful open-source weight model. In our internal Z.ai code benchmark test, its performance has improved by 50% compared to GLM-5.2. Additionally, during the scale deployment after training, its network capacity development speed exceeded our expectations.

On the CyberGym platform, GLM-5.3 is in a leading position in vulnerability discovery, with the most significant improvement in the exploitation chain's backend— in the exploitation benchmark test, its performance is more than twice that of GLM-5.2. After the completion of security assessment and enhancement work, we will disclose the weights two weeks after the release.