Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-08-14 · source-backed
Z.ai released GLM-5.3 August 14 on the same base model as 5.2, attributing a claimed 50% coding gain entirely to scaled post-training, ranking first among open-source models on Terminal Bench 3.0 and Agents' Last Exam. On CyberGym it scored 84.5%, up from 5.2's 77.2%. The company says vulnerability-discovery training environments produced an unintended emergent ability to reason across full multi-stage exploitation chains rather than isolated bugs, and is withholding weights roughly two weeks for safety hardening. Reportedly the first time a Chinese lab has cited emergent offensive capability as a release-delay reason. Z.ai says its models have found 2,436 vulnerabilities across 269 open-source projects since 5.2, 1,097 rated critical or high.
Each link below shares sources, entities, or timing with this story.
Shared entities / Shared topic / Earlier coverage
Both cover August, Chinese, GLM; overlapping topics (august, model); earlier August coverage from 2026-08-07.
Shared entities / Shared topic / Earlier coverage / Tension
Both cover Chinese, GLM; overlapping topics (agent, model); earlier Chinese coverage from 2026-05-05.
Shared entities / Shared topic / Earlier coverage
Both cover August, GLM; overlapping topics (actually, agent, august); earlier August coverage from 2026-08-04.
Both cover Chinese, GLM; overlapping topics (actually, aren, model); earlier Chinese coverage from 2026-07-08.
Shared entities / Shared topic
Both cover August, GLM; overlapping topics (agent, august, model).
Shared entities / Shared topic / Earlier coverage
Both cover August, Chinese; overlapping topics (agent, model); earlier August coverage from 2026-08-05.
Both cover August, GLM; overlapping topics (august, capability); earlier August coverage from 2026-08-05.
Both cover August, Terminal Bench; overlapping topics (agent, model); earlier August coverage from 2026-08-02.