Base44 launches proprietary Base1 model to strengthen defensibility for vibe coding platform

Base44, the Wix-owned vibe coding platform, is rolling out its own proprietary large language model, Base1, to power app creation through natural language. The move is a direct response to the growing debate over whether AI startups can remain defensible when built on top of third-party models. By owning its model, Base44 aims to improve latency, reduce inference costs, and build a vertically integrated stack that combines data, distribution, and infrastructure.

What happened

Base44 announced the rollout of its own AI model, Base1, trained on tens of millions of real user interactions on the platform. Wix acquired Base44 for $80 million about a year ago when the startup was only six months old. Since then, Base44 has reportedly surpassed $100 million in annual recurring revenue, though competitor Lovable claimed $500 million ARR. The company has been growing headcount while its parent Wix recently announced a 20% workforce reduction.

Why AI builders should care

For AI builders and platform owners, the question of defensibility is central. Jonathan Userovici, a general partner at Headline, identifies data, distribution, and tech stack as the three pillars of defensibility for AI startups. Base44’s move to own its model aligns with the data and infrastructure aspects, giving it direct control over compute and inference spend. This contrasts with competitors like Lovable, which rely on external LLMs, and frontier labs like Anthropic that are moving into vibe coding with Claude Code.

Inference costs have become a meaningful part of the equation. Enterprise customers are increasingly demanding cost-conscious orchestration across models to avoid skyrocketing expenses while maintaining performance. Base44’s founder Maor Shlomo argues that owning the model allows for optimizations that frontier models like Opus cannot match for specialized app-building use cases.

Practical implications

Base44 expects that owning Base1 will deliver lower latency, lower cost, and better efficiency for its users. The company also anticipates a structurally stronger margin profile over time by controlling compute and inference spend directly. For developers using Base44, this could mean faster app generation and more predictable pricing, though the model is still rolling out and has not yet proven its performance against frontier models.

The broader trend points toward specialized, vertically integrated solutions. As Userovici notes, applied AI companies are unlikely to become frontier labs en masse, but those with enough scale and data can benefit from owning their stack. The legal tech startup Harvey, which abandoned plans to train its own model, serves as a cautionary example that this path is not for everyone.

Caveats

Base1 is only just rolling out, and its performance relative to frontier models like Opus remains unproven. Competitors such as Lovable, Cursor, and Grok are also investing in their own models or leveraging parent company resources. Frontier AI labs continue to improve general-purpose models, and Shlomo himself expects that other players with sufficient scale will train their own models. The cost of developing and maintaining a proprietary LLM is significant, and the payoff in margins may take time to materialize. Details on Base1’s architecture, benchmarks, and pricing have not been disclosed.

FAQs

What is Base44's Base1 model and how does it work?

Base1 is a proprietary large language model developed by Base44 for its vibe coding platform. It is trained on tens of millions of real user interactions from the platform and is designed to help users build apps and websites through natural language commands.

Why is Base44 building its own AI model instead of using third-party options?

Base44 founder Maor Shlomo says owning the model allows for optimizations on latency, cost, and efficiency that are not possible with general-purpose frontier models. The move is also a defensibility play, giving Base44 direct control over compute and inference spend to improve margins over time.

How might Base44's in-house model affect latency and costs for developers?

Base44 expects that owning Base1 will result in faster and cheaper app generation compared to using frontier models like Opus. The company also anticipates a structurally stronger margin profile by controlling inference costs directly, though the model is still in early rollout.

How does Base44's approach to defensibility compare to other AI startups?

Industry voices highlight data, distribution, and tech stack as key defensibility pillars. Base44’s vertical integration through model ownership aligns with data and infrastructure control. Competitors like Lovable rely on external LLMs, while frontier labs like Anthropic are moving into vibe coding with their own models. The legal AI startup Harvey abandoned its own model effort, showing the approach is not universally successful.