Exploring AI Model Performance in UI/UX and Coding: Insights from a Global Survey

In recent months, I conducted an extensive research project to evaluate the capabilities of various AI models in the realms of user interface/user experience (UI/UX) design and coding. This initiative involved collecting data from over 6,000 participants worldwide, providing valuable insights into which AI tools excel and which fall short.

A Crowdsourced Benchmark for AI-Generated Content

To facilitate this analysis, I developed a crowdsourced benchmarking platform—DesignArena.ai—where users can generate and compare a wide range of digital assets, including websites, games, 3D models, and data visualizations across different AI models. Since launching, nearly 5,000 users have engaged with the platform, submitting close to 4,000 votes. This collaborative effort has yielded a comprehensive overview of AI performance in creative and technical tasks.

Key Findings from the User Comparisons

• Top Performers in Coding and Design:
  Among the evaluated models, Claude and DeepSeek stood out as leaders. The user preference leaned heavily towards Claude Opus, which ranked highest on the leaderboard. Notably, DeepSeek models, especially version 0, gained popularity for their robust website generation capabilities. Interestingly, Grok emerged as a surprising dark horse—fast and capable, despite being less publicly recognized. However, it’s worth noting that DeepSeek models tend to operate slowly, making Claude a more practical choice for interface development.

• The Underrated yet Powerful Grok 3:
  While it doesn’t garner as much online attention—perhaps partly due to its association with Elon Musk—Grok 3 performed impressively. It consistently ranks within the top five and offers significant speed advantages over other models, making it a noteworthy option for developers.

• Variability in Gemini 2.5-Pro:
  Gemini 2.5-Pro has shown mixed results. User feedback indicates that its UI/UX outputs can be hit or miss; sometimes delivering well-designed interfaces, other times producing subpar applications. While it demonstrates competence in coding business logic, its consistency in design remains an area for improvement.

• Performance Gaps in OpenAI and Meta Models:
  OpenAI’s GPT models exhibit average performance—neither leading nor lagging significantly. In contrast