Z.ai releases GLM-5.3 with 50% coding gain

Z.ai released GLM-5.3 on the same 743-billion-parameter base as GLM-5.2, claiming post-training alone moved Terminal-Bench 3.0 from 4.6% to 28.3% and its internal coding score up 50%. The technical release report says CyberGym reached 84.5%, while downloadable weights will follow in roughly two weeks after safety hardening; teams can use the API now, but should wait for weights if residency and vendor-control requirements matter. The release is a useful reminder for open-weight builders buying control at a premium: post-training and harness design can change the model’s operating value without changing its base checkpoint.