Sakhanda Wire
NVDA $152.58 -2.28% MSFT $506.11 -2.29% GOOGL $279.81 -1.38% META $619.07 -1.17% AMZN $256.55 -0.47%
← Back to the news

Z.ai releases GLM-5.3 with post-training gains

Z.ai released GLM-5.3 on August 14, 2026, using the unchanged 743B GLM-5.2 base model. The company attributes the reported improvements entirely to expanded post-training: more long-horizon task environments, a wider range of environment types and longer training. The model is available through the Z.ai API, GLM Coding Plan and ZCode, while its weights are expected roughly two weeks after launch, following safety evaluation and hardening.

Coding results improved most on extended tasks. Terminal-Bench 3.0 rose from 4.6 to 28.3, DeepSWE v1.1 from 46.2 to 66.9, and Agents’ Last Exam (CLI) from 23.8 to 28.5. Z.ai also reports a 50% improvement over GLM-5.2 on its internal Z.ai Code Bench, with 31.4% at about 50,000 output tokens per task. The company says GLM-5.3 trails GPT-5.6 Sol and Fable 5 on several harder public coding tests.

Cybersecurity performance also increased. CyberGym reached 84.5%, compared with 77.2% for GLM-5.2, while ExploitBench rose from 24.4% to 54.4%. Z.ai reports that GLM-5.3 completed 105 ExploitGym tasks in two hours and 130 in six hours. All figures are vendor-reported.

Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.

Originally published by MarkTechPost on

Read the original on MarkTechPost ↗

The reporting and the image belong to MarkTechPost and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news