Vibe coding platform Base44 launches own model as AI startups seek defensibility

Basic44the vibe coding platform that Wix bought for $80 million a year ago – when the company was just six months old and had a team of eight – has started rolling out its own AI model to support its users in creating natural language apps.

The move comes as debate has intensified in AI circles over whether boundary models are best suited for all use cases. A related question is whether companies built on someone else’s models are really defensible in the long run. The latest move from Bay Area-based Base44 speaks to both.

Although the modified LLM is only just being rolled out, Base44 hopes it will eventually outperform frontier models. According to the founder, Maor Shlomo, “Training and owning the model is part of it [our] The entire stack gives us many more optimizations in terms of latency, cost and efficiency.”

At first glance, this could be a way to stay ahead of competitors like Swedish startup Lovable, which reached unicorn status in its Series A round last summer and that depends on external LLMs. However, Shlomo expects others will train their own models – “at least the players that have gained enough scale and speed to have enough data.”

According to Jonathan Userovici, general partner at VC firm Headline – whose portfolio includes AI companies like Mistral AI, but not Base44 – data is one of the three key ingredients of AI startups’ resilience, along with distribution and tech stack.

The result is that players with strong brands are now using their data and infrastructure to increase their resilience, and Base44 fits that pattern. The company says the first version of its LLM, Base1, was developed and trained on a dataset generated from “tens of millions of real user interactions on the platform.”

This data set will continue to grow with the company; but so do his rivals.’ The bigger competition may not be vibe-coding startups at all, but instead comes from groundbreaking AI labs that are moving closer to Base44’s home base. Cursor and Grok’s parent company xAI both now belong to SpaceX, and Claude Code has become a “vibe-coding” player in his own right.

This gives Anthropic and other foundational AI providers access to data and feedback loops they can use to improve app building models, but Shlomo thinks specialization gives Base44 an edge. “Models are making progress, but they will remain very generic in what they can do,” he predicted.

Userovici, in turn, warned against underestimating boundary models, citing the example of legal technology startup Harvey, which abandoned plans to train its own model. He doesn’t expect applied AI companies to become frontier labs en masse, but puts Base44’s move in a broader context – one in which inference costs have become a meaningful part of the equation.

According to Userovici, this cost pressure has led to changes that business customers are now demanding. “They don’t necessarily see one [return on investment] so when using the latest models for all use cases, a whole infrastructure is 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 for most use cases.”

Enterprise companies are still a minority among the vibe coding platforms’ audience, but they represent a growing share of platform revenues, and users of all sizes are starting to express concerns about the costs of using AI. Base44’s decision to develop its own LLM stemmed from several factors, but cost savings is likely one of the benefits.

“We want to get a model that is more in line with what we think is right, that will be more optimized for what we users like in terms of the results we get, and that will ultimately be faster and cheaper for customers than using frontier models like Opus,” Shlomo said.

As for Base44 itself, the cost reduction is not so clear. In a press release, the company explained that “ownership of the Base44 model provides direct control over compute and inference spend, which is expected to result in a structurally stronger margin profile over time.”

Even with a deferred payout, improved margins would be good news for Base44’s parent company, which recently announced it Lay off 20% of the workforce. In contrast, Base44’s workforce has grown since the acquisition – and has announced that it has passed $100 million in annual recurring revenue a few months ago.

That’s still less than Lovable, which said earlier this month it had reached $500 million in ARR. But Shlomo is betting that the “enormous technical effort” to develop Base1 will cement Base44’s positioning as the “only vertically integrated vibe-coding application” – which, in Userovici’s terms, means a player that owns its distribution, data and infrastructure all at once.