We blind tested Grok 4.7, Kimi K3 and GPT-6 for design
1.1万 次观看 · 2026-09-25
UI 设计
We gave Grok 4.7, Kimi K3 and GPT-6 Astra the same brief and hid their names. The job was a tribute website for the architect Luis Barragán, and we judged every site before we knew who made it.
Everyone has a favourite AI model, and once you get comfortable with 1 tool it's hard to test the others fairly. So we built a blind model comparison app. It uses OpenRouter with your own API key, runs up to 6 models at once, and shuffles them so you can't tell which is which. We gave each model a written brief, a 4,000 character fact file on Barragán's 10 best known works, and our GSAP motion library at 02ui.com/motion, then compared speed, cost and design.
What you'll see:
• How the blind test works, and why hiding the model name changes your judgement
• The 6 part prompt structure we use for design briefs
• Real cost and time for each model, from 1 cent to 49 cents per site
• What happens when a run fails, and why you still pay for it
• Why every site worked and none of them felt like Barragán
Time stamps:
0:00 Grok 4.7 vs Astra vs Kimi K3: blind AI design test
0:22 Why there is no single best AI model for design
1:32 Our blind test app: compare AI models with OpenRouter
2:04 Why blind testing removes brand bias
3:25 API credits vs subscription: why testing costs more
6:13 The brief: a Luis Barragán tribute website
8:05 Adding GSAP animations from zerotoui.com/motion
11:00 How we structure prompts for better AI design
14:43 Round 1: judging the websites blind
19:17 Round 1 reveal: Astra 7/10, Kimi K3 6/10, Grok 4.7 errors
24:09 Round 2: the same prompt, run again
24:56 Why the AI missed the brief: human taste and priorities
26:01 How we would design it: paper, pen and collage first
30:59 Cost per model: Astra vs Grok 4.7 vs Kimi K3
If you want to see round 2 with Mobbin references, subscribe and it'll turn up in your feed.
→ 02UI channel: youtube.com/@zerotoui
Follow us on :
https://www.youtube.com/@zerotoui
https://www.instagram.com/zerotoui
https://www.tiktok.com/@zerotoui
https://www.linkedin.com/company/zerotoui/
https://x.com/zerotoui
Lets connect :
https://www.linkedin.com/in/muratbayral/
https://www.linkedin.com/in/canhoskan/
Work with us:
https://www.miragraph.com/
https://www.linkedin.com/company/miragraph/
https://www.linkedin.com/company/zerotoui/
Recently added videos:
🔥 How to build a design system from scratch with Claude Code
https://youtu.be/tzn8AAJI6jw
🟠 I tried to break my Claude Max plan with Fable 5.1
https://youtu.be/hLLyX-d-R9Y
🟠 Your own AI agent team on your own API key
https://youtu.be/E5sYb0bVEDs
🟠 I turned my blog into a museum you walk through
https://youtu.be/gCcy0FBzw5I
🟠 We built the same invoice app 2 ways with Claude Code
https://youtu.be/W5RTbAoBOPo
🔥 Claude Design vs Lovable: One Prompt, Same Brief
https://youtu.be/o4jIKc_DIoM
02UI website: https://02ui.com
Affiliate links: https://02ui.com/affiliate
AD | Affiliate link: (Mobbin) https://mobbin.com/zerotoui
Get 20% discount on the annual subscription
02UI Lab teaches the full journey from first file to shipped product.
Join waiting list: https://02ui.com/lab
#Figma #Claude #DesignSystems
Everyone has a favourite AI model, and once you get comfortable with 1 tool it's hard to test the others fairly. So we built a blind model comparison app. It uses OpenRouter with your own API key, runs up to 6 models at once, and shuffles them so you can't tell which is which. We gave each model a written brief, a 4,000 character fact file on Barragán's 10 best known works, and our GSAP motion library at 02ui.com/motion, then compared speed, cost and design.
What you'll see:
• How the blind test works, and why hiding the model name changes your judgement
• The 6 part prompt structure we use for design briefs
• Real cost and time for each model, from 1 cent to 49 cents per site
• What happens when a run fails, and why you still pay for it
• Why every site worked and none of them felt like Barragán
Time stamps:
0:00 Grok 4.7 vs Astra vs Kimi K3: blind AI design test
0:22 Why there is no single best AI model for design
1:32 Our blind test app: compare AI models with OpenRouter
2:04 Why blind testing removes brand bias
3:25 API credits vs subscription: why testing costs more
6:13 The brief: a Luis Barragán tribute website
8:05 Adding GSAP animations from zerotoui.com/motion
11:00 How we structure prompts for better AI design
14:43 Round 1: judging the websites blind
19:17 Round 1 reveal: Astra 7/10, Kimi K3 6/10, Grok 4.7 errors
24:09 Round 2: the same prompt, run again
24:56 Why the AI missed the brief: human taste and priorities
26:01 How we would design it: paper, pen and collage first
30:59 Cost per model: Astra vs Grok 4.7 vs Kimi K3
If you want to see round 2 with Mobbin references, subscribe and it'll turn up in your feed.
→ 02UI channel: youtube.com/@zerotoui
Follow us on :
https://www.youtube.com/@zerotoui
https://www.instagram.com/zerotoui
https://www.tiktok.com/@zerotoui
https://www.linkedin.com/company/zerotoui/
https://x.com/zerotoui
Lets connect :
https://www.linkedin.com/in/muratbayral/
https://www.linkedin.com/in/canhoskan/
Work with us:
https://www.miragraph.com/
https://www.linkedin.com/company/miragraph/
https://www.linkedin.com/company/zerotoui/
Recently added videos:
🔥 How to build a design system from scratch with Claude Code
https://youtu.be/tzn8AAJI6jw
🟠 I tried to break my Claude Max plan with Fable 5.1
https://youtu.be/hLLyX-d-R9Y
🟠 Your own AI agent team on your own API key
https://youtu.be/E5sYb0bVEDs
🟠 I turned my blog into a museum you walk through
https://youtu.be/gCcy0FBzw5I
🟠 We built the same invoice app 2 ways with Claude Code
https://youtu.be/W5RTbAoBOPo
🔥 Claude Design vs Lovable: One Prompt, Same Brief
https://youtu.be/o4jIKc_DIoM
02UI website: https://02ui.com
Affiliate links: https://02ui.com/affiliate
AD | Affiliate link: (Mobbin) https://mobbin.com/zerotoui
Get 20% discount on the annual subscription
02UI Lab teaches the full journey from first file to shipped product.
Join waiting list: https://02ui.com/lab
#Figma #Claude #DesignSystems