Z.ai's GLM-5.3 ties Kimi K3 atop open models but holds the weights two weeks
GLM-5.3 scores 60 on the Artificial Analysis Intelligence Index — level with Kimi K3, seven points over GLM-5.2 — and climbs to 1,770 Elo on GDPval-AA v2 agentic tasks, second only to Claude Opus 5 at 1,855, at $0.68 per task against Kimi K3's $0.84. It is live on Z.ai's API, but the open weights are held roughly two weeks while Z.ai tightens controls and restricts full access to selected security partners, citing the model's skill at finding security vulnerabilities. A Chinese lab staging an open release on security-capability grounds is the new part; the price pressure on Kimi K3 lands now, the weights do not.
Source: the-decoder.com ↗
GLM-5.3 is so effective at detecting security vulnerabilities that it is first strengthening controls and restricting full access to select security partners.
Why this matters
- → Chinese model now matches Kimi K3 at the top; undercuts it 19% on price.
- → Agentic capabilities leap 246 Elo points; second only to Claude Opus 5.
- → Security-capability grounds for delaying weights release signals new release-strategy frontier.