Vibe Coding Platform Base44 Launches Its Own AI Model in Bid for Defensibility

Base44, the vibe-coding platform Wix acquired for $80 million a year ago, has begun rolling out its own AI model — making it, by its own claim, the first app-creation platform to deploy a proprietary large language model in production rather than relying entirely on frontier models from OpenAI, Anthropic, or Google. What Base44 actually built The new model, called Base1, isn't trained from scratch — it's a fine-tune of an existing open-source model, built in collaboration with Wix's machine learning team. What makes it notable is the training method: Base44 ran the model repeatedly against real platform tasks — building and editing applications — scoring the outputs and feeding that signal back through reinforcement learning. The training data itself came from tens of millions of real user interactions on the platform, a dataset only a company with Base44's scale could have assembled. Founder and CEO Maor Shlomo says training and owning the model lets the company optimize directly for latency, cost, and efficiency in ways that renting someone else's frontier model can't match — though he's been candid that he doesn't expect Base1 alone to reshape the competitive landscape. He's framed it instead as the first release in an ongoing series, with larger models and deeper product integration planned. Why this is happening now The launch lands amid a broader industry conversation about whether AI applications built entirely on top of someone else's models can ever be truly defensible. According to Jonathan Userovici, a general partner at VC firm Headline, defensibility for AI startups rests on three things: data, distribution, and tech stack. Companies with a strong user base — like Base44 — are increasingly leaning into the data and infrastructure legs of that stool, since the model layer itself is rented and replicable by competitors. Cost is the other driver. Inference spending has become one of the largest line items for AI-native companies, and enterprise customers in particular are pushing back on paying frontier-model prices for tasks that don't need frontier-level capability. That pressure is fueling a wider shift toward routing and orchestration — picking cheaper, narrower models for routine tasks and reserving expensive frontier models for the cases that actually need them. The competitive picture is more complicated than "us vs. other vibe-coding startups" Base44 isn't just racing rivals like Lovable, the Swedish vibe-coding startup that reportedly hit $500 million in annual recurring revenue this month — well ahead of Base44's own growth. The bigger threat may come from frontier labs moving directly into Base44's territory: Cursor and xAI now both sit under the SpaceX umbrella, and Anthropic's Claude Code has become a vibe-coding competitor in its own right. If the companies building the frontier models start competing on Base44's home turf, owning a fine-tuned model trained on your own users' data becomes less of a luxury and more of a survival strategy. Why it matters This is a useful early data point in a question the whole AI application layer is going to have to answer: what happens to companies whose entire product is a thin layer on top of someone else's model, once the model providers themselves start building competing products? Base44's bet — invest engineering effort now to own data, infrastructure, and distribution together — is a hedge against that exact scenario. Shlomo's framing of Base44 as the "only vertically integrated vibe-coding application" is also a signal to investors and enterprise customers that the company sees its survival tied to control over the full stack, not just a clever prompt layer. Whether that bet pays off depends on whether a fine-tuned narrow model can keep pace as frontier models keep improving — a race Base44 has now explicitly entered.