Zhipu's GLM-5.2 tops global AI design benchmark

Zhipu AI's GLM-5.2 model just scored first place in the Design Arena single-round HTML web design test, beating Claude Fable 5 and others. Design Arena is a global platform that uses crowdsourced blind tests to judge AI design quality, seen as a key benchmark for aesthetics and real-world design.

GLM-5.2 jumped 5 spots from its previous version GLM-5.1. Its inference cost is only $1.40/$4.40 per million tokens, much cheaper than Fable 5's $10/$50. The model efficiently uses libraries like chart.js and three.js, boosting win rates by 6% in those sessions.

It uses TailwindCSS in 91% of sessions and font-awesome in 51%, improving win rates by 1.2% through better interaction design. In comparison, Fable 5 only uses TailwindCSS in 57% of sessions. GLM-5.2 excels in layout, typography, and animation, especially when building web pages with external CDN images.