What is Gestalt AI? The Model Topping the Web Design Blind Test

Gestalt AI is an anonymous model showcased by Contra Labs in a web design evaluation. It topped the leaderboard reviewed on November 20, 2026, with 1,931 Elo, ahead of the human reference works and Claude Fable 5.2. No public trial access has been found yet. To understand this evaluation, you can start by opening the official gallery to see the design briefs given to the competing models. Official release · Online leaderboard · Works gallery

What is Gestalt AI? Can you try it now?

Contra Labs marks it as Gestalt (stealth). Public materials describe its performance on design tasks, but the development team, underlying model, and access method remain undisclosed. As of November 20, 2026, this research also found no official API, pricing, or release date. If you see a download page for a multimodal model named Gestalt, first verify the publisher. GeWu Lab has a research project of the same name, and its public page includes its own paper, model, and demo. Contra Labs release · GeWu Lab project page

What do the web pages made by Gestalt look like?

In a spot check of four public projects—Bad Move, Bandstand, Carafe, and NovaraAI—no pages explicitly labeled as Gestalt were found. For now, you can view case studies from the same prompts in the official gallery, but Gestalt's own layout, visual style, and interaction details still lack verifiable samples. Official gallery · Bad Move · Bandstand · Carafe · NovaraAI

For example, Bad Move is a crowdfunding page for a board game publisher. The brief requires users to be able to select preset amounts, initiate recurring support, and see a wall of backers on the page. The project page lists versions such as the human reference and Claude Fable 5.2. Opening this project lets you compare the brief with the public works and observe how the same crowdfunding requirements translate into pages and user flows. Bad Move project page

How does Gestalt compare with Claude and human designs?

Online leaderboard rankModel or workElo
1Gestalt (stealth)1,931
2Human reference works1,739
3Claude Fable 5.21,713

The reviewers see two works under the same brief and choose the one they prefer with the authors' identities hidden. According to the November 18 release, Gestalt was directly compared with human reference works 28 times, with 20 reviewer selections for Gestalt. The release records Gestalt's score as 1,928; the table uses the online leaderboard data as of November 20. Evaluation methodology · Release · Online leaderboard

If you care about specific details like typography, brand fit, and usability, you can also check the itemized design critiques. Gestalt scored 83, while Claude Fable 5.2 and Claude Opus 5.5 both scored 84. The blind selection reflects reviewers' preference for the overall work, while the critique table checks design details item by item. Both scores come from this evaluation; deployment performance, real conversions, and code maintenance costs still need separate verification. Scoring rubric

Want to see it yourself? Where to go?

To see what pages can be made from the same brief, open the project from the official gallery, read the brief first, then compare the human reference and public model versions. To check Gestalt's relative performance, look at the online leaderboard. To learn what this model preview disclosed, read the Gestalt section in the Contra Labs release. Official gallery · Online leaderboard · Contra Labs release