Base44 Launches Base 1, Its Own Proprietary AI Model
Base44Base44 launched Base 1, its first in-house AI model, making it the first vibe-coding platform to ship a proprietary model of its own. Fine-tuned from an open-source base on tens of millions of real user building interactions, Base 1 is designed to fix the generic "AI slop" look that frontier models tend to produce and to give Base44 more control over latency, cost, and output quality. The launch is widely read as a strategic defensibility move as vibe-coding platforms face commoditization pressure from general-purpose frontier labs.
Sources & Mentions
3 external resources covering this update
Base44 Bets on Owning Its Own Model
Base44 introduced Base 1, a proprietary AI model built specifically for the platform's app-building workflows. The launch, announced by founder and CEO Maor Shlomo, makes Base44 the first app-creation ("vibe coding") platform to ship a model it trained and owns rather than relying exclusively on third-party frontier labs like Anthropic or OpenAI.
Trained on Base44's Own Building Data
Base 1 is a fine-tune of an existing open-source base model rather than something trained from scratch. What makes it distinctive is the training data: tens of millions of real user interactions collected from people actually building apps on Base44. According to Shlomo, this domain-specific dataset gives Base 1 an edge in understanding how users describe, iterate on, and refine software in practice, rather than relying only on generic web-scale training data. The team also applied reinforcement learning to push the model toward generating designs that look fresh and distinct rather than repeating the same visual patterns turn after turn.
Solving the "AI Slop" Problem
One of the most visible complaints about vibe-coded apps is that outputs from frontier models tend to look interchangeable, the same layout patterns, the same component choices, the same generic aesthetic. Base 1 was built in part to address this directly, positioning itself as a model tuned for variety and originality in UI generation rather than raw general-purpose intelligence.
Strategic Context
The move comes as Base44, acquired by Wix for over $80 million in 2025, has scaled to roughly 2 million users and a reported $150 million annualized revenue run rate. Owning a proprietary model is framed as a bet on margin control and defensibility: training and hosting a model in-house lets Base44 optimize latency, cost, and efficiency in ways that pure resale of third-party model access cannot. It also signals Base44's ambition to compete on its own AI stack rather than being purely a wrapper around frontier labs, a distinction increasingly important as the vibe-coding category matures and competitors converge on similar feature sets.
In Base44's model picker, Base 1 is positioned as a well-rounded, general-purpose choice suited for everyday building and conversation, sitting alongside higher-end options for more complex or multi-step work.