Base44 just made a bet that could reshape how vibe coding platforms compete. The company, which Wix acquired for $80 million barely a year ago, has started rolling out its own AI model to help users build apps with plain natural language. According to TechCrunch AI, the Bay Area startup is doing this to answer a question that’s been gnawing at the whole industry: can a business built on top of someone else’s models actually survive long-term?
The new model is called Base1. Founder Maor Shlomo says owning the model end to end gives the team room to optimize in ways renting a frontier model never could. “Training and owning the model as part of [our] entire stack allows us a lot more optimizations on latency, cost, and efficiency,” he told TechCrunch AI.
What stands out here is the timing. The industry is openly debating whether frontier models are the right tool for every job, and whether companies that lease intelligence from labs like Anthropic or OpenAI have any real moat. Base44 is planting a flag on the other side of that argument.
What Base1 actually is
Here’s what the article lays out about the model and the thinking behind it:
- Trained on the platform’s own data. Base44 says the first version of Base1 was built on a dataset drawn from “tens of millions of real user interactions on the platform.”
- Built for speed and cost. Shlomo wants a model that’s “faster and cheaper for customers eventually than using the frontier models like Opus.”
- A margin play for the parent company. Base44 says owning the model gives it “direct control over compute and inference spend,” which it expects to produce “a structurally stronger margin profile over time.”
- Still early. The custom LLM is only just rolling out, and the company openly hopes it will eventually outperform frontier models, not on day one.
Why this matters
This is significant because it reframes what defensibility means for applied AI startups. Jonathan Userovici, a general partner at VC firm Headline, told TechCrunch AI that data is one of three key ingredients of a moat, alongside distribution and tech stack. Companies with strong brands are now leaning into their own data and infrastructure, and Base44 fits that pattern neatly.
The competitive picture is worth understanding. Swedish rival Lovable hit unicorn status last summer and still relies on external LLMs. Base44 thinks owning its stack puts daylight between the two. But Shlomo expects imitation: he believes other players “that have gotten enough scale and velocity to have enough data” will eventually train their own models too.
The real threat may come from above
The bigger competition might not be other vibe coding startups at all. Cursor and Grok’s parent xAI now both sit under SpaceX, and Claude Code has become a serious vibe coding player on its own. That gives Anthropic and other foundational labs direct access to app-creation data and feedback loops. Shlomo’s counterargument is specialization. “Models are progressing, but they’ll stay very general in what they can do,” he predicted.
Userovici offered a useful caution. He pointed to legal tech startup Harvey, which abandoned its own model plans, and warned against underestimating frontier labs. He doesn’t expect applied AI companies to become frontier labs in bulk. Instead, he frames the moment around inference costs, which have become a meaningful line item. Enterprise customers, he says, “don’t necessarily see a [return on investment] when using the latest models for all use cases,” so the market is building orchestration layers to route the right model to the right task without letting costs explode.
The numbers behind the bet
Base44 passed $100 million in annual recurring revenue a few months ago and has been adding headcount since the acquisition. That’s still well behind Lovable, which reported $500 million in ARR earlier this month. And improved margins would help Base44’s parent, which recently announced it would lay off 20% of its workforce.
Shlomo’s wager is that the “huge engineering effort” behind Base1 cements Base44 as the only vertically integrated vibe coding application, owning its distribution, data, and infrastructure at once. Whether specialization beats raw frontier scale is the question the next year will answer. For the full breakdown, the original reporting is at TechCrunch AI.