Benchmarks
LABS
Benchmarks for AI on real UI work
How frontier models copy, build and design UI.
Nov 17, 2026 · 2 benchmarks live · 22 models














Benchmarks2 live·Leaderboard →
Copy Bench
Models rebuild a web component from a short screen recording.
GPT-6.2 Astra92.4
GPT-6.2 Sol88.4
GPT-6 Astra86.7
GPT-6.1 Sol82.2
Turing UI Bench
AI-made hero sections shown in the 21st feed next to designers' work.
Top designer4.7%
GPT-6 Astra3.4%
GPT-6.1 Sol3.3%
Grok 4.72.8%
Median designer2.7%
Key results
Copy Bench


A browser check adds up to 12 points
92.4
+3.3
GPT-6.2 Astra
88.4
+4.3
GPT-6.2 Sol
86.7
+9.8
GPT-6 Astra
82.2
+6.6
GPT-6.1 Sol
69.0
+6.6
Claude Fable 5.2
67.2
+10.2
Gemini 4 Carbon
Score / 100 · faded part = after the browser check
Read the article →Turing UI BenchThe best designer still beats every model
4.7%
Top designer
3.4%
GPT-6 Astra
3.3%
GPT-6.1 Sol
2.8%
Grok 4.7
2.7%
Claude Sonnet 5.5
2.7%
Median designer
Pick rate in the feed · striped = designers
Read the article →Research
4 studies · built on 21st data
Nov 16, 2026Reward model
Can a model learn developer taste in UI?

Nov 2, 2026Turing UI Bench
AI-made vs human-made hero sections, tested on real developers

Nov 2, 2026Taste
AI models predict which UI components developers choose, up to 70% of the time

Oct 27, 2026Animation
Can AI models rebuild an animation from screenshots?
21st Labs · partners
Test your model on real UI work
Private runs before release. Custom tasks from the 21st catalog.








