Base44 wants to help users ditch the cookie-cutter look of AI-coded websites with its new AI model.
The San Francisco-based vibe coding startup announced on Monday that it had trained and released its own large language model called Base 1. Before giving the prompt to the platform to build an app, users can now select Base 1 from a selection of AI models, including Claude’s Opus 4.8, Fable 5, and OpenAI’s GPT-5.5.
Shlomo told Business Insider that part of the reason his company developed its own model was to combat the telltale generic look that many vibe-coded products have.
He said the problem with using frontier models for making websites is that “everybody feels like they’re getting the same UI when they’re coding with the general models.”
Base44 was acquired by Wix last June for $80 million, making it part of Wix’s website-building arsenal. Shlomo said Wix has a large team of designers, which generates a lot of data for its model to train on.
The Base44 team will conduct “reinforcement learning” on the new model, Shlomo said, which involves prompting it to keep generating designs that look new and unique.
While Base 1 is “not yet there,” Shlomo said he aims for it to “create something that looks uniquely different” each time it generates a user interface. He said the team started working on Base 1 about six months ago, but had breakthroughs in the last few weeks, allowing them to release the model sooner than planned.
He said generating unique designs will set Base44 apart from competitors. Startups like Lovable, Replit, and Cursor compete with Base44.
Many UI/UX experts and design gurus have warned about the AI slop look on vibe-coded websites and apps. Paul Bakaus, the CEO of AI design startup Impeccable, said in a June interview with Andreessen Horowitz that telltale signs of AI-coded products include beige or tinted backgrounds and sans-serif fonts.
Bakaus likened it to an “algorithmic Uniqlo or Ikea,” a design that’s not bad, but not unique.
Shlomo told Business Insider that frontier models must be generically good at everything, from poetry to coding.
“And we think that if we take a model and we squeeze its ability to be really, really good at one use case, then we have a shot,” he said.
Read the full article here
