Base44 rolls out custom LLM to support its vibe-coding platform

Base44, a vibe-coding platform acquired by Wix for $80 million in 2025, has started rolling out Base1, its own custom AI model designed to support users in creating applications through natural language instructions. The company positions this as a step toward outperforming frontier models by optimizing for latency, cost, and efficiency. Base44’s founder, Maor Shlomo, stated that owning the model allows for "a lot more optimizations" within the company’s stack.

The first iteration of Base1 was trained on a dataset derived from "tens of millions of real user interactions on the platform," according to the company. This dataset is expected to grow as Base44 scales, though it will compete with similar datasets from rivals. The move reflects a broader industry trend where companies seek defensibility through data, distribution, and infrastructure, rather than relying solely on third-party models.

The decision to develop Base1 was driven by multiple factors, including cost reduction and alignment with user needs. Shlomo noted that the custom model would be "more optimized to what we see users like in terms of the results we’re getting" and "faster and cheaper for customers eventually than using frontier models like Opus." The company also highlighted that ownership of the model provides "direct control over compute and inference spend," which is expected to improve margins over time.

While Base44’s custom model is still in its early stages, the company’s leadership views it as a key differentiator in a competitive landscape. Shlomo predicted that "models are progressing, but they’ll stay very general in what they can do," suggesting that specialization could give Base44 an advantage in vibe coding. However, the company acknowledges competition not only from other vibe-coding startups but also from frontier AI labs like Anthropic, whose Claude Code has entered the vibe-coding space.

The broader context includes rising enterprise demand for cost-effective AI solutions. Jonathan Userovici, a general partner at Headline (which invests in Mistral AI but not Base44), framed Base44’s move within a trend where enterprises seek to optimize model selection to balance performance and cost. Userovici cautioned that while some companies may train their own models, others—like the legal tech startup Harvey—have opted against it, highlighting the trade-offs involved.

Base44’s financial performance contrasts with its parent company’s recent workforce reductions. Base44 reported passing $100 million in annual recurring revenue, though this remains below Lovable’s reported $500 million in ARR. Despite the upfront costs of developing Base1, the company views the "huge engineering effort" as a strategic investment in becoming the "only vertically integrated vibe-coding application," owning its distribution, data, and infrastructure.