Base44, the vibe coding platform Wix acquired for $80 million just a year ago, is now rolling out its own AI model. The Tel Aviv-based startup wants to prove that specialization beats generalization when it comes to building apps with natural language.
The company’s first LLM, called Base1, was trained on tens of millions of real user interactions from the platform. Founder Maor Shlomo says owning the model lets them optimize for latency, cost, and efficiency in ways that relying on frontier models doesn’t allow.
It’s a bet on defensibility. The AI startup world has been wrestling with a tough question: if your product runs on someone else’s model, do you actually have a moat? Base44 thinks the answer is no — and that owning your own model, data, and infrastructure is the only way to build something lasting.
But the competition isn’t standing still. Cursor and xAI (Grok’s parent) are now both owned by SpaceX. Claude Code has become a serious vibe-coding player. Frontier labs are getting closer to Base44’s home turf with every model release.
Jonathan Userovici, a general partner at VC firm Headline, points out that data alone isn’t enough. Distribution and tech stack matter too. And he cautions against underestimating frontier models — citing legal tech startup Harvey, which abandoned plans to train its own model.
Still, the cost argument is real. Enterprise customers are pushing back on inference costs, and Base44 thinks a purpose-built model can deliver similar performance at a fraction of the price. Whether that holds up at scale is the real test.
