Zhipu officially announced the launch of GLM-5.3 API today. In the authoritative Artificial Analysis Comprehensive Intelligence Index (AA Index), this model scored 60 points, successfully entering the global cutting-edge model capability range. It not only matches closed-source flagship models such as Claude Fable5 and GPT-5.6Sol, but also ranks alongside Kimi K3 as the top open-source model.

image.png

In terms of core capabilities, GLM-5.3 demonstrates strong overall performance, especially excelling in complex coding, defensive cybersecurity, and long-range tasks. According to official information, GLM-5.3 uses the same base model as its predecessor GLM-5.2, and its significant performance improvement mainly comes from deep optimization during the post-training phase.

image.png

Notably, this model significantly lowers the access threshold for advanced AI while achieving the same level of intelligence. In Artificial Analysis' "Intelligence-Cost" evaluation, GLM-5.3 has the lowest single-task cost among cutting-edge flagship models, placing it in the most attractive range of high intelligence and low cost. This pushes the Pareto frontier of cutting-edge models significantly, aiming to achieve higher intelligence performance with lower cost per unit.

Currently, GLM-5.3 has been fully integrated into coding platforms such as ZCode and has officially been incorporated into the GLM Coding Plan. The API is fully available for calls, and the pricing remains consistent with its predecessor GLM-5.2. Meanwhile, the official has confirmed that the model's weights will be released as open source on next Friday, further empowering real-world production and work scenarios for developers, enterprises, and knowledge workers.