Zhipu AI Model GLM 5.2 Tops Design Arena HTML Design Benchmark
Zhipu AI's GLM 5.2 model has claimed the top position on the Design Arena single-round HTML web design (non-agent) leaderboard. It surpassed Anthropic's Claude Fable 5, Opus 4.6, and Opus 4.7, achieving first place.
The Design Arena platform evaluates AI-generated design quality through crowdsourced blind-tests. It is widely considered an industry-relevant benchmark for aesthetics, and practical design capability.
GLM 5.2 offers a substantial cost advantage with inference pricing at USD 1.40 per million input tokens, and USD 4.40 per million output tokens. This significantly undercuts Claude Fable 5's pricing of USD 10 for input, and USD 50 for output.
The model demonstrates efficient utilisation of third-party libraries, including chart.js and three.js. Sessions using these libraries saw a 6.0 percentage point win rate improvement.
GLM 5.2 employs TailwindCSS in 91% of its sessions, and font-awesome in 51%. Carefully designed interactions boosted win rates by an additional 1.2 percentage points.
In comparison, Fable 5 uses TailwindCSS in only 57% of sessions, which may contribute to its lower performance. GLM 5.2 also shows significantly improved layout capabilities, skilfully incorporating external CDN images.
It delivers outstanding performance in typography, visual layout, and animation effects. This achievement marks a significant milestone for Chinese AI models in the creative domain.
While previous Chinese models have excelled in language understanding, mathematics, and coding, Design Arena measures visual design aesthetics. GLM 5.2's top ranking demonstrates China's advancing AI capabilities across diverse cognitive domains.
GLM 5.2 represents the latest iteration in Zhipu AI's rapid development cycle. Its open-source release has been welcomed by the developer community, expanding access to advanced AI design capabilities.
* Zhipu AI's GLM 5.2 model achieved first place on Design Arena's HTML web design benchmark.
* GLM 5.2 surpassed Anthropic's Claude Fable 5, Opus 4.6, and Opus 4.7.
* The model offers a significant cost advantage over Claude Fable 5, priced at USD 1.40 per million input tokens and USD 4.40 per million output tokens.
Source: Pandaily
