The flagship model GLM 5.3 from SmartModel will be open-sourced on August 28th at 10:00 AM.

Perceive Beating AI News Flash: The flagship model GLM-5.3 will be open-sourced on August 28 at 10 a.m. Beijing time.

GLM-5.3 actually went online with its Coding Plan as early as August 14, followed by the opening of the API. Tomorrow, the model weights can finally be downloaded, delayed by a full two weeks for a somewhat special reason. Perceive stated that during the post-training process, the network's cybersecurity capabilities improved at a rate "beyond expectations." Not only is it better at finding vulnerabilities but it has also started to learn to plan a complete exploit chain across multiple steps. Therefore, the official team performed a security assessment and reinforcement before releasing the weights.

In official evaluations, GLM-5.3 scored 84.5% in vulnerability discovery tests using CyberGym, slightly higher than Mythos 5 at 83.8% and GPT-5.6 Sol at 83.6%. In terms of attack exploitation, ExploitBench increased from 24.4% in GLM-5.2 to 54.4%, although still significantly behind Mythos 5 at 78.0%. Perceive also mentioned that the model has discovered 2436 vulnerabilities in 269 open-source projects, with 1097 classified as critical or high-risk.

GLM-5.3 continues to use the same base model as GLM-5.2, with all these improvements coming from post-training.