Base44 rolls out custom model to beat frontier costs

Base44 rolls – Base44, the vibe-coding platform Wix acquired for $80 million a year ago, is now rolling out its own AI model, Base1, built from tens of millions of real user interactions. The company says owning and training its model lets it optimize latency, cost and effic

On a platform built for speed—typing natural language and watching an app take shape—time and money are suddenly part of the product.

Base44. the vibe coding company Wix acquired for $80 million just one year ago. has started rolling out its own AI model to help users create apps using natural language. The move lands in the middle of a debate that’s been heating up inside AI circles: whether “frontier” models are the right fit for every use case. and whether businesses built on top of someone else’s model can defend themselves long-term.

Base44, based in the Bay Area, is betting that the answer is yes—if it owns the model.

Its founder. Maor Shlomo. says training and owning the model as part of Base44’s stack gives the company room to optimize key factors like latency. cost. and efficiency. Even as Base44’s custom large language model is only just rolling out. the ambition is explicit: it should eventually outperform frontier models.

Shlomo’s logic is simple. At first. the step looks like a way to stay ahead of competitors such as Swedish startup Lovable. which reached unicorn status in its Series A round last summer and relies on external LLMs. But Shlomo expects others to follow the same path once they have the data and scale to justify it. “Others will train their own models. ” he argues—“at least the players that have gotten enough scale and velocity to have enough data.”.

That emphasis on data is echoed by Jonathan Userovici, a general partner at VC firm Headline. Userovici’s portfolio includes AI companies like Mistral AI, though not Base44, and he points to a three-part recipe for defensibility in applied AI: data, distribution, and the tech stack.

Base44 says its first iteration, Base1, was developed and trained on a dataset generated from “tens of millions of real user interactions on the platform.” The company expects that dataset to keep growing as usage grows—while competitors do too.

And the competition isn’t confined to vibe-coding startups.

Userovici and Shlomo both frame the shift against a backdrop where frontier labs are moving closer to Base44’s territory. Cursor and xAI’s Grok—xAI now belongs to SpaceX—are part of that pressure. Claude Code has also become a vibe-coding player.

There’s another tension here. one that matters beyond any single app builder: foundational AI providers can access data and feedback loops when they power tools like these. Shlomo argues specialization can still create an edge. “Models are progressing, but they’ll stay very general in what they can do,” he predicted.

Still, not every applied AI company is willing to make the leap from using a frontier model to training its own.

Userovici cautioned against underestimating frontier models, pointing to legal tech startup Harvey, which abandoned plans to train its own model. He also links Base44’s move to a shift in what enterprise customers are demanding now.

Inference costs have become a real line item. not a background detail. and enterprise buyers want orchestration and optimization—systems designed to select the right models without letting spending “skyrocket.” In his words. enterprise customers “don’t necessarily see a [return on investment] when using the latest models for all use cases. ” so infrastructure is being set up to keep performance similar while controlling costs across most use cases.

For Base44, cost reduction is one part of the story. Shlomo says the company wants a model aligned with how Base44 believes users should get results—optimized so it’s “faster and cheaper for customers eventually than using the frontier models like Opus.”

But Base44’s own press release makes the financial mechanism look more direct than the typical pitch.

The company explained that “ownership of the model gives Base44 direct control over compute and inference spend,” which it expects will produce a “structurally stronger margin profile over time.”

That matters because Base44 is tied to Wix—an acquisition that, at the time, made Base44’s scale feel almost unreal. Wix bought the company when it was barely six months old, with a team of eight, in a deal valued at $80 million.

Since then, Base44 has been growing. Even as Wix recently announced it would lay off 20% of its workforce, Base44 says it has increased headcount since the acquisition. It also announced it had passed $100 million in annual recurring revenue a few months ago.

Lovable, meanwhile, has been moving faster in raw numbers. It said it hit $500 million in ARR earlier this month.

Base44 may still be behind on scale, but Shlomo is arguing that the “huge engineering effort” behind Base1 could change what it is: less of an app builder powered by others’ models, more of a vertically integrated player.

In Userovici’s framing, Base44’s target is to be “the only vertically integrated vibe-coding application”—a company that owns distribution, data, and infrastructure all at once.

In a market where the next model release can reshuffle costs overnight, Base44 is trying to build a different kind of advantage—one not only based on how good the app generator is, but on who controls the bill.

Base44 Wix acquisition vibe coding AI model Base1 custom LLM inference costs Opus Lovable Cursor xAI Claude Code Harvey Headline defensibility