SECURITY Signal 388
Z.ai claims GLM-5.3 scores 84.5% on CyberGym, ahead of Mythos 5, and restricts sensitive cyber functions to verified users
↗
Z.ai announced GLM-5.3, an open-source model that scores 84.5% on CyberGym versus Mythos 5's 83.8%, and will limit its most sensitive cybersecurity functions to verified users.
For engineers, GLM-5.3 offers near-frontier cybersecurity capability in an open-weight model, but the most sensitive functions require verified access, so integration plans must account for that gate. The close CyberGym score suggests open models are closing the gap with restricted ones, which may change how teams evaluate model options for security tasks.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
GLM-5.3 scores 84.5% on CyberGym, slightly above Mythos 5's 83.8%.
Z.ai will restrict the model's most sensitive cybersecurity functions to verified users.
GLM-5.3 uses the same base model as GLM-5.2, with gains from post-training, and weights are due in two weeks.
THE CLUSTER