Base44, the vibe-coding platform acquired by Wix for $80 million a year ago, has begun deploying its own large language model to help users create apps using natural language. At the time of the acquisition the company was only six months old and had a team of eight. Base44 is based in the Bay Area.
Why this matters
The move addresses two ongoing debates in AI: whether frontier models are the best fit for all use cases, and whether businesses built on other companies’ models can remain defensible long-term. By developing its own model, Base44 aims to exert more control over performance, latency and costs.
The new model: Base1
Base44 says the first iteration, called Base1, was developed and trained on a dataset originating from “tens of millions of real user interactions” on the platform. Founder Maor Shlomo has argued that training and owning the model as part of Base44’s stack enables more optimizations around latency, cost and efficiency.
Costs, defensibility and customer demands
Owning the model, the company notes, gives direct control over compute and inference spend and is expected to result in a structurally stronger margin profile over time. Cost pressure is becoming a meaningful factor for customers: enterprise buyers in particular are pushing for infrastructure that can orchestrate and optimize which models are used so cost doesn’t balloon while preserving similar performance.
Investor perspective and competitive context
Jonathan Userovici, a general partner at VC firm Headline (whose portfolio includes AI firms such as Mistral AI), identifies data, distribution and tech stack as three core ingredients of defensibility for AI startups. He also cautions not to underestimate frontier models — citing examples like legal-tech startup Harvey, which abandoned plans to train its own model — and notes that applied AI companies do not typically transform into frontier labs en masse.
A bigger competitive threat may come from frontier AI labs encroaching on Base44’s territory. Cursor and Grok’s parent company xAI now both belong to SpaceX, and Claude Code has also entered the vibe-coding space. Those foundational providers can access data and feedback loops to improve models for app creation; nonetheless, Shlomo believes specialization—being tuned to vibe-coding use cases—will be a competitive advantage.
Financial and organizational effects
Base44’s parent company recently announced a plan to lay off 20% of its workforce, while Base44 itself has been hiring since the acquisition and said it passed $100 million in annual recurring revenue (ARR) a few months ago. By comparison, Lovable — a Swedish rival that relies on external LLMs — said earlier this month it reached $500 million in ARR. Base44 hopes the substantial engineering effort behind Base1 will help it remain the only vertically integrated vibe-coding application, owning distribution, data and infrastructure simultaneously.
Conclusion
Base44’s in-house LLM is a strategic bet to improve latency, reduce cost and strengthen margins by leveraging proprietary data and infrastructure. The initiative aligns with a broader industry trend toward model ownership and optimization, even as competition from frontier AI providers and other specialized startups continues to intensify.
