Base44 Unveils Unique AI Model Amid Rise of Defensible Startups in the Coding Arena

Base44, a platform specialized in vibe coding which was acquired by Wix for $80 million just a year ago—when it was still in its infancy with a mere team of eight—has begun to launch its own AI model aimed at assisting users in app creation through natural language.

This initiative emerges amidst ongoing debates in the AI community regarding the suitability of frontier models for various applications. An accompanying concern is whether businesses built using third-party models can maintain long-term viability. Base44, hailing from the Bay Area, is making strides in this area.

Although its custom large language model (LLM) is just beginning to roll out, Base44 envisions surpassing existing frontier models in performance. Founder Maor Shlomo stated, “Training and controlling the model within our entire stack allows us significant optimizations related to latency, cost, and efficiency.”

This strategy may help Base44 stay competitive against rivals like Lovable, a Swedish startup that achieved unicorn status during its Series A funding last summer and depends on external LLMs. Still, Shlomo anticipates that other companies will also develop their own models as they attain sufficient scale and data velocity.

Jonathan Userovici, a general partner at the venture capital firm Headline, noted that data is one of three crucial elements for the defensibility of AI startups, alongside distribution and technology stack.

As a result, companies with established brands are exploiting their data and infrastructure to enhance their competitive edge, and Base44 exemplifies this trend. The company has reported that the initial version of its LLM, Base1, was developed using a dataset compiled from “tens of millions of genuine user interactions on the platform.”

This dataset is expected to expand continually, although its competitors will similarly grow their data resources. In terms of competition, the greater challenge might not just be from vibe-coding startups, but also from frontier AI labs encroaching on Base44’s home region—Cursor and Grok’s parent company xAI are now part of SpaceX, while Claude Code is growing its presence in vibe coding.

This situation allows established AI providers like Anthropic to gather data and create feedback loops that could enhance app development models. However, Shlomo believes that Base44’s specialization gives it a competitive advantage. “While models are advancing, they will remain quite general in functionality,” he stated.

Userovici cautioned against dismissing frontier models, providing the example of legal tech startup Harvey, which scrapped its plan to create a proprietary model. He doesn’t foresee a sweeping trend of applied AI firms transforming into frontier labs but positions Base44’s development efforts within a wider framework—one that acknowledges the significant cost implications that businesses are now grappling with.

Userovici highlighted that enterprises are now demanding changes due to cost pressures. “They don’t always perceive a [return on investment] when employing the latest models across all applications. As a result, an entire infrastructure is being established to orchestrate and optimize model selection, ensuring costs remain manageable while maintaining consistent performance.”

While enterprise users currently make up a smaller segment of the audience for vibe coding platforms, they are becoming an increasingly important source of income, and concerns regarding AI costs are rising among users of all sizes. Base44’s initiative to create its own LLM was influenced by several factors, with cost reduction being a key advantage.

“We aim to develop a model that aligns more closely with what we define as optimal, providing results that resonate with user preferences, ultimately being faster and more cost-effective than relying on frontier models like Opus,” Shlomo stated.

For Base44, the path to cost reduction isn’t entirely straightforward. A company statement indicated that “having ownership of the model grants Base44 direct oversight over computing and inference expenditures, anticipated to result in a more robust margin structure over time.”

Even though potential returns may take time, improved margins would be positive news for Base44’s parent company, which has recently announced a 20% workforce reduction. Conversely, Base44 has been expanding its staff since the acquisition and announced it surpassed $100 million in annual recurring revenue a few months ago.

While this figure lags behind Lovable’s reported $500 million in annual recurring revenue, Shlomo believes that the significant engineering investment in developing Base1 will solidify Base44’s status as the singular vertically integrated vibe-coding platform—encompassing distribution, data, and infrastructure under one roof.