GLM-5.3 Released: MIT Licensed, Focused on Agent Coding and Security

baseten · x · 2026-08-29

Z.AI released GLM-5.3, a 743B MoE model under the MIT license. The model achieves significant improvements in long-horizon agentic capabilities and cybersecurity analysis through scaled post-training, with Terminal-Bench 3.0 scores jumping from 4.6% to 28.3%. It supports three thinking effort levels (low, high, max), designed for production-scale coding workflows requiring sustained planning and verification.

Related event: Z.ai Releases GLM-5.3 Open Weights, Touting Agentic Coding and Cyber Defense(7 posts)→

Original post →

More from coding & agent

coding & agent channel →