Base44, the innovative vibe coding platform acquired by Wix for $80 million just a year ago—when it was a nascent company with a mere eight employees—has begun the rollout of its proprietary AI model designed to assist users in app development using natural language.
This initiative occurs amid intensifying discussions in the AI community regarding the suitability of frontier models across various use cases. Furthermore, a key point of contention is whether businesses that rely on external models have a sustainable competitive advantage over the long term. Base44, headquartered in the Bay Area, is addressing both of these concerns.
While its custom large language model (LLM) is still in its early stages, the company aspires for it to eventually surpass existing frontier models. Founder Maor Shlomo stated, “by training and owning the model as part of [our] complete stack, we have greater opportunities for optimizing latency, cost, and overall efficiency.”
This strategy might provide Base44 with an edge over competitors such as the Swedish startup Lovable, which reached unicorn valuation during its Series A funding last summer and utilizes external LLMs. Shlomo also anticipates that other companies with sufficient scale and data will start training their own models.
Jonathan Userovici, a general partner at the VC firm Headline—which invests in AI companies like Mistral AI but not Base44—asserts that data, along with distribution and tech stack, are critical components for the defensibility of AI startups.
As a result, major players are increasingly capitalizing on their data and infrastructure to bolster their market position, and Base44 is aligning itself with this trend. The company reports that its initial LLM, Base1, was developed and trained using a dataset compiled from "tens of millions of genuine user interactions on the platform."
This dataset will continue to expand alongside Base44, even as competitors ramp up their offerings. The more significant competition might not stem solely from other vibe-coding startups, but rather from frontier AI labs that are encroaching on Base44’s territory. Companies like Cursor and Grok’s parent, xAI, now part of SpaceX, and Claude Code are both entering the vibe coding space.
These developments grant established AI providers like Anthropic access to valuable data and feedback loops, which can enhance app creation models. Nevertheless, Shlomo maintains that Base44's specialized approach offers an advantage. He believes that while models are advancing, they will remain largely generalized in their applications.
Userovici offered a note of caution against underestimating frontier models, pointing to the legal tech firm Harvey, which decided against developing its own model. He does not foresee a mass transition of applied AI companies into frontier labs but positions Base44's initiative within a larger framework—in which the costs associated with inference are becoming increasingly significant.
Userovici explains that this cost sensitivity has prompted changes that enterprise clients now demand. “They don’t always perceive a [return on investment] when implementing the latest models for every application, leading to the establishment of infrastructures designed to optimize and orchestrate the selection of appropriate models without inflating costs and while maintaining performance across various use cases.”
While enterprise clients represent only a fraction of users on vibe coding platforms, their contribution to platform revenues is increasing, and users of all sizes are starting to raise concerns about AI usage costs. Base44’s strategy to develop its own LLM is influenced by several factors, with cost management likely being a primary advantage.
Shlomo remarked, “We aim to create a model that aligns more closely with our vision, optimizes what users prefer in terms of results, and ultimately proves to be faster and cheaper for customers than existing frontier models such as Opus.”
However, for Base44 itself, the path to cost reduction is not entirely straightforward. The company communicated in a press release that “ownership of the model grants Base44 direct control over compute and inference expenses, which is expected to lead to a stronger margin profile over time.”
Even if the financial benefits are not immediate, improved margins would be beneficial for Wix, which recently announced a 20% workforce reduction. In contrast, Base44 has seen growth in its workforce since the acquisition and reported that it surpassed $100 million in annual recurring revenue a few months ago.
Although this figure still trails behind Lovable, which claimed $500 million in annual recurring revenue earlier this month, Shlomo believes that the significant engineering effort put into developing Base1 will solidify Base44's position as the only vertically integrated vibe-coding application—encompassing its distribution, data, and infrastructure all at once, as Userovici describes it.
