Z.ai’s GLM-5.3 model has secured the top position on the Artificial Analysis Intelligence Index for open systems by scoring 60 points. This result matches Kimi K3 while placing the new version seven points ahead of its predecessor, GLM-5.2. The primary performance increase occurs in agentic tasks where the model achieved an Elo score of 1,770 on the GDPval-AA v2 benchmark. This figure ranks it second globally behind only Claude Opus 5. Cost analysis suggests the model is priced at $0.68 per task, which is 19 percent cheaper than Kimi K3 despite being more expensive than GLM-5.2.
The release of open weights is postponed by approximately two weeks due to internal security concerns. Z.ai states that the model effectively detects security vulnerabilities, prompting the company to strengthen controls before granting full access. The model remains available via the Z.ai API for now.
* Performance gain of 246 points on agentic benchmarks
* Price of $0.68 per task
* Delayed open weights due to security risk detection
