NewsBite analysis · AI, Programming · New AI Model
The Decoder
GLM-5.3 tops the open-model rankings and undercuts rivals on price, but its release is delayed
Z.ai has announced that its GLM-5.3 model has topped the open-model rankings, tying with Kimi K3. With significant performance improvements, it boasts a competitive price per task; however, the release of its open weights has been postponed by two weeks. This delay is due to its advanced capabilities in detecting security vulnerabilities, prompting the company to enhance security controls before wider access is granted.
MonitorConfidence: MediumShort term horizon
Published 19 Aug 2026, 14:10 UTC1 min read
Visit original article ↗Lead image — The Decoder
Summary
TL;DR
- GLM-5.3 scores 60 on the AI model ranking index.
- Ranks tied with Kimi K3 but outperforms its predecessor GLM-5.2.
- Elo score increases significantly on GDPval-AA benchmark.
- Cost-effective pricing at $0.68 per task, cheaper than Kimi K3.
- Release delayed by two weeks for security enhancements.
Interesting facts
- GLM-5.3 shows a notable improvement in agentic tasks.
- The release delay indicates a proactive approach to security by Z.ai.
Coverage gaps
- Details on security measures that are being implemented.