Base44 is moving deeper into the infrastructure behind its vibe-coding platform. The Tel Aviv company, acquired by Wix for $80 million just one year ago, has started rolling out its own AI model to help users build apps through natural language.
The new model, called Base1, is more than a product update. It is a signal about where applied AI companies think the next layer of competition may come from: not only better interfaces, but control over data, costs, latency, and the model stack itself.
Why Base44 is training its own model
Base44 was barely six months old and had a team of eight when Wix acquired it. Since then, the company has kept expanding, while the broader AI market has been asking a difficult question: can startups built on top of someone else’s models remain defensible over time?
Base44’s answer is to own more of the system. Founder Maor Shlomo said that "training and owning the model as part of [our] entire stack allows us a lot more optimizations on latency, cost, and efficiency."
That logic is especially relevant for vibe coding, where users expect natural-language prompts to become working apps quickly. If a platform can tune its own model around the exact interactions happening inside its product, it may be able to make the experience faster, cheaper, or more closely aligned with user expectations.
The first version of Base1 was developed and trained on a dataset generated from "tens of millions of real user interactions on the platform." That data is central to the company’s argument. As more people use Base44, the dataset can continue to expand, giving the company more feedback from the exact use case it serves.
Defensibility now means more than a polished interface
Jonathan Userovici, a general partner at VC firm Headline, framed defensibility for AI startups around three ingredients: data, distribution, and tech stack. Headline’s portfolio includes AI companies like Mistral AI, but not Base44.
Base44 is trying to bring those ingredients together. It has a user-facing app creation product, usage data from its platform, and now its own model layer. Shlomo is betting that this combination will support Base44’s position as the "only vertically integrated vibe-coding application."
This does not mean every AI startup will train its own model. Shlomo expects the move to be limited to companies with enough scale and momentum to have enough data. In his view, the players with sufficient scale and velocity are the ones most likely to follow a similar path.
The strategy also reflects a broader shift among AI companies with strong brands. They are increasingly using their own data and infrastructure to strengthen their position rather than relying only on access to external frontier models.
The competition is not only other vibe-coding startups
Base44’s move may look like a direct response to rivals such as Swedish startup Lovable, which reached unicorn status in its Series A round last summer and relies on external LLMs. Lovable also said it hit $500 million in ARR earlier this month.
Base44 has been growing too. It announced it had passed $150 million in annual recurring revenue in May, two months after crossing $100 million in ARR. Even so, the larger competitive question may involve foundational AI providers rather than only vibe-coding platforms.
Frontier AI labs are moving closer to app creation. Cursor and Grok’s parent company xAI now both belong to SpaceX, and Claude Code has become a vibe-coding player in its own right. That gives companies such as Anthropic and other foundational AI providers access to feedback loops that can improve models for building apps.
Shlomo’s counterargument is specialization. He said, "Models are progressing, but they’ll stay very general in what they can do." Base44’s premise is that a model trained around its own app-building workflow can serve that specific job better than a general-purpose model in at least some cases.
Cost pressure is becoming part of the model debate
The decision to build Base1 also sits inside a growing concern about inference costs. Userovici cautioned against underestimating frontier models and pointed to legal tech startup Harvey, which abandoned plans to train its own model.
Still, he sees a wider shift happening. Enterprise customers do not always see enough return from using the newest models for every task. As he put it, "They don’t necessarily see a [return on investment] when using the latest models for all use cases, so an entire infrastructure is being set up to do orchestration and optimization to select the right models for them so that costs don’t skyrocket while maintaining the same or similar performance across the majority of use cases."
Enterprise companies are still a minority among vibe-coding platform users, but they represent a growing share of platform revenue. Users of different sizes are also beginning to raise concerns about the cost of AI usage.
For customers, Base44 wants Base1 to become more aligned with what users want, more optimized around preferred results, and eventually faster and cheaper than using frontier models like Opus. Shlomo described the goal this way: "We want to get a model that is going to be more aligned to what we think is the right thing, is going to be more optimized to what we see users like in terms of the results we’re getting, and is going to be faster and cheaper for customers eventually than using the frontier models like Opus."
The payoff may take time
For Base44 itself, the cost benefit is not immediate or guaranteed. The company said in a press release that "ownership of the model gives Base44 direct control over compute and inference spend, expected to result in a structurally stronger margin profile over time."
That matters for Wix, which recently announced it would lay off 20% of its workforce. Base44, by contrast, has continued to grow in headcount since the acquisition.
The larger takeaway is that Base44 is treating its AI model as part of the product’s strategic foundation, not just a backend choice. In vibe coding, the companies that own the interface, the data, and the infrastructure may have more room to shape performance, price, and user experience as the market matures.