Base44, the vibe coding platform owned by Wix, began rolling out its own large language model in 2026, marking a shift away from reliance on external AI providers such as Anthropic’s Claude. The company, based in the Bay Area, says the model — called Base1 — was trained on data generated from tens of millions of user interactions on its platform.
Wix acquired Base44 for $80 million roughly one year ago, when the startup was six months old and employed just eight people. Since then, Base44 has grown its headcount and recently reported passing $100 million in annual recurring revenue.
Founder Maor Shlomo said developing and owning the model gives Base44 “a lot more optimizations on latency, cost, and efficiency.” The company described the move as positioning Base44 as the “only vertically integrated vibe-coding application” — one that controls its distribution, data, and infrastructure simultaneously. Shlomo added that he expects Base1 to eventually outperform frontier models for the platform’s specific use cases, arguing that general-purpose models “will stay very general in what they can do.”
Cost pressure is a key driver. Base44 said ownership of the model gives it “direct control over compute and inference spend, expected to result in a structurally stronger margin profile over time” — a meaningful consideration given that parent company Wix recently announced layoffs of 20% of its workforce.
Jonathan Userovici, a general partner at VC firm Headline, framed the broader context: enterprise customers are increasingly questioning the return on investment of using the most powerful frontier models for every task, spurring demand for infrastructure that selects cheaper models without sacrificing performance.
The competitive landscape extends beyond rival vibe coding startups like Lovable — which reported $500 million in ARR and reached unicorn status — to frontier AI labs moving into the space directly. Claude Code has become a vibe coding player in its own right, and Cursor and xAI now both fall under SpaceX, giving those organizations access to data feedback loops for app creation. Userovici cautioned against underestimating frontier models, noting that legal tech startup Harvey abandoned plans to train its own model.
Source: TechCrunch