GLM-5.3
From the source
z.aiPublisher’s summary
Z.ai's next model reuses the GLM-5.2 base and puts every gain into post-training — weights held back roughly two weeks this time, after training produced cybersecurity capabilities the team didn't plan for.