How well does AI
build Arabic?
Every month, free AI models build ten Arabic screens on their own, then again with Dualkit. dualkit check scores every build against 34 rules.
Next run in
The first run lands on 1 November.
Six free models, ten screens, three runs each. The results and every raw file go up here the same day.
Read the promptsHow the Bench works
Ten screens
A payment summary, a one-time code, an address form, a profile, an order history, a checkout, a support chat, a booking, settings and transactions. Every prompt is public.
Three runs, two ways
Each model builds each screen three times on its own, then three times with the Dualkit skills loaded. The middle run counts.
Scored by dualkit check
The same 34 rules you can run with npx dualkit check. A build that leaves out a part the prompt asked for scores 0.
A small sample
Free models only, run once a month. A model that fails shows as didn’t run, never as a guess.