Val Town built a conference site in 3 minutes, while v0 took 15: testing six AI services
Developer Pete Millspaugh gave six AI services the same request: build a one-page cheat sheet for his conference talk, with links and audience comments. Val Town produced a working version in 3 minutes, but the interface came out cluttered, while v0 spent about 15 minutes and cut out mid-generation several times. According to the author, Claude Artifacts made the best design. When evaluating a prototype, look beyond speed to whether forms and login actually work.
AI-processed from Habr AI; edited by Hamidun News
If you give six AI services the same request, "make me a website," the results vary widely. Developer Pete Millspaugh tested this on a real task: he needed a one-page cheat sheet for the attendees of his talk. Val Town was the fastest, delivering a working site in 3 minutes, but its interface came out cluttered. v0 performed worst: it needed about 15 minutes, and the model stopped working several times, so the generation had to be restarted. Below we look at how the others did and what follows from it.
What exactly was requested
The talk is called "Your corner of the internet" and is about building personal websites. Millspaugh gave it as a workshop at Startup Cincy Week. Attendees needed a single page on the domain yourcorneroftheinternet.com, purchased through Cloudflare. The page had to serve as a cheat sheet: links and materials useful during and after the presentation. It also needed comments, so that attendees and anyone else could ask questions, share their own sites and recommend resources.
Millspaugh planned to fill the site with content himself, so the services were only asked to build the "skeleton": the page structure and working functions. The request said nothing about styling, so each service chose its own design. Lovable, v0, Replit, ChatGPT Sites, Claude Artifacts and Val Town all received the same request.
How each service performed
Val Town delivered a working version in 3 minutes. It was the fastest result, but the author found the interface overloaded.
ChatGPT Sites on the GPT-6.1 Sol model finished in 4 minutes 17 seconds. The service published the site and explained how to connect a custom domain. Comments worked without authentication or moderation. That is convenient for a conference page, but an open form without checks will eventually attract spam.
Lovable took 4 minutes 55 seconds. It asked the author about design preferences and rules for comments. However, authentication in the discussions still did not work. Clarifying questions during the process do not guarantee a working result.
Claude Artifacts took a different approach. Instead of a complex structure built on a web framework, it produced several HTML, JavaScript and SQL files. This option was simpler, and the author liked Claude's design the most. For a small one-page task, that makes sense: fewer dependencies, easier to understand what is inside, and simple to edit by hand.
v0, as noted, spent about 15 minutes and stopped several times mid-generation.
Replit also took part in the comparison. The retelling of the experiment contains no detailed remarks about its result, so its performance cannot be assessed from this material.
What follows from this
First: speed and quality do not always go together. Val Town was the fastest, but its interface came out cluttered. Claude Artifacts won the author's favor on design, although it is not named among the leaders on time.
Second: "done" means different things for different services. ChatGPT Sites handled publishing and the domain, while Lovable's authentication broke. Before showing a site to people, check not only how it looks but also whether the forms, login and comments actually work.
Third: a simple task does not need a complex stack. When the model writes plain HTML, JS and SQL instead of using a framework, the result is easier to read and modify.
There are also honest limitations. This is a single experiment with one prompt, and one person judged the design. It cannot be called a systematic benchmark, and the services update quickly, so tomorrow's ranking may differ.
Millspaugh summed up the results with mixed feelings. He is optimistic about today's website-building capabilities: the idea that anyone can tailor software to their needs seems more achievable than ever. He is also pessimistic: a site made of AI slop, which the person barely thought about or did not think about at all, is often worse than having no site at all. The practical takeaway is simple: AI produces the scaffolding in a few minutes, while the content, verification and responsibility for the result remain with the human.
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
Key Questions Answered
What exactly was requested?
The talk is called "Your corner of the internet" and is about building personal websites. Millspaugh gave it as a workshop at Startup Cincy Week. Attendees needed a single page on the domain yourcorneroftheinternet.com, purchased through Cloudflare. The page had to serve as a cheat sheet: links and materials useful during and after the presentation. It also needed comments, so that attendees and anyone else could ask questions, share their own sites and recommend resources.
How each service performed?
Val Town delivered a working version in 3 minutes. It was the fastest result, but the author found the interface overloaded. ChatGPT Sites on the GPT-6.1 Sol model finished in 4 minutes 17 seconds. The service published the site and explained how to connect a custom domain. Comments worked without authentication or moderation. That is convenient for a conference page, but an open form without checks will eventually attract spam. Lovable took 4 minutes 55 seconds.
What follows from this?
First: speed and quality do not always go together. Val Town was the fastest, but its interface came out cluttered. Claude Artifacts won the author's favor on design, although it is not named among the leaders on time. Second: "done" means different things for different services. ChatGPT Sites handled publishing and the domain, while Lovable's authentication broke. Before showing a site to people, check not only how it looks but also whether the forms, login and comments actually work.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.