Put the latest frontier models through the classic "pelican on a bicycle" SVG benchmark (pedal physics + inverse kinematics).
Ran the exact same prompt across GPT-6 Astra, Claude Opus 5, Gemini 3.7 Flash, and GLM 5.3 Flash side-by-side. Clip below: - GPT-6 Astra: Nailed the
Claude Opus 5 is crazy for web design.
Prompt ↓
A professional designer directly broke the defense after trying GPT-6 Astra
“I originally thought that my design sense would not lose to Fable 5, Opus 5, GPT-5.6… Until I used Astra. " He directly asked Astra to build a complete Design System in Figma, starting from the visual identity, and then to the details of the final interface. The point is not that "AI can draw UI". Instead it begins to understand:
Runway just turned the UI into a video.
Runway’s Solaris is its first Interface World Model: instead of HTML/DOM, it generates an interactive interface as continuous 720p video. An LLM handles reasoning while an adapted Gen-4.5 renders responses to user actions in real time.
You have reached the end of the archive
All of Claude Opus 5