Base44 launches custom AI model to boost efficiency

Base44, a startup based in Tel Aviv, has started rolling out its own large language model (LLM), called Base1. The model is designed to support users in “vibe coding”—the process of creating software applications using natural language. To build the model, the company trained Base1 on a dataset generated from tens of millions of real user interactions on its platform.

The launch follows Base44’s acquisition by Wix for $80 million last year. At the time of the acquisition, the startup was barely six months old and had a team of eight. According to founder Maor Shlomo, “training and owning the model as part of [our] entire stack allows us a lot more optimizations on latency, cost, and efficiency.” Shlomo hopes the custom LLM will eventually outperform frontier models—the large, general-purpose AI models typically provided by external developers.

This vertical integration comes amid intense competition in the natural language development space. Base44 is competing directly with rivals like Swedish startup Lovable, though their financial scales differ significantly:

• Base44: Passed $100 million in annual recurring revenue (ARR) a few months ago.
• Lovable: Hit $500 million in ARR earlier this month, having reached unicorn status last summer.

While Lovable currently relies on external LLMs, Shlomo expects that at least the players with enough scale and velocity will eventually train their own models. Jonathan Userovici, a general partner at venture capital firm Headline, noted that data is one of three key ingredients of defensibility for AI startups, alongside distribution and the technology stack.

Developing a custom LLM is likely to reduce costs, giving Base44 direct control over compute and inference—the process of running a trained AI model. Improved margins would be good news for parent company Wix, which recently announced it would lay off 20% of its workforce. However, the broader industry is facing challenges with high inference costs. Userovici cautioned that many enterprise customers do not necessarily see a return on investment when using the latest models for all use cases. This has led some firms to focus on orchestration and optimization to select the right models rather than building proprietary ones, similar to legal tech startup Harvey, which abandoned its own model-training plans.

Why it matters

Base44’s move to develop its own LLM reflects a broader trend among AI startups seeking defensibility by owning their data and infrastructure to reduce costs and improve margins.