Zhipu AI released its next-generation base model, GLM-5.3, via API in the early hours of Monday morning.

The model excels in complex coding, defensive cybersecurity, and long-horizon tasks. It achieved a score of 60 on the Artificial Analysis Intelligence Index, placing it firmly within the capability range of the world's leading frontier models. This puts it on par with closed-source flagships like Claude Fable 5 and GPT-5.6 Sol, while also tying with Kimi K3 for the top spot among open-source models.

Currently, the frontier of AI development is shifting away from simply competing over larger parameter counts and toward delivering effective intelligence at a lower unit cost. GLM-5.3 reportedly reaches frontier-level performance using a smaller parameter scale, achieving higher parameter efficiency and lower inference costs. It secures the lowest per-task cost among leading flagship models, significantly advancing the "intelligence-to-cost" Pareto frontier. This progress is expected to accelerate the deployment and large-scale adoption of AI applications.

The performance leap in GLM-5.3 was not achieved by merely increasing parameter size. It builds on the same base model as its predecessor, GLM-5.2, with the gains primarily stemming from enhancements in post-training. This demonstrates that advancing frontier capabilities does not necessarily require larger parameter counts or greater computational investment.

The model is now integrated into official products such as ZCode and the GLM Coding Plan. Its API pricing remains consistent with GLM-5.2, and the model weights are scheduled to be open-sourced next Friday.