Base44 launches Base1, its first custom AI model trained on user interactions

Wix-owned vibe-coding platform Base44 has begun rolling out Base1, its first proprietary large language model, trained on tens of millions of real user interactions from the platform. TechCrunch reported the launch on June 29, 2026.

Base44 was acquired by Wix for $80 million in June 2025 and has since grown to $150 million in annual recurring revenue. It operates in a segment where competitor Lovable has reached $500 million in ARR—a comparison that illustrates how large and competitive the vibe-coding market has become in a short period.

Founder Maor Shlomo says Base1 is optimized for latency, cost, and efficiency on its specific task domain: helping users create applications through natural language. The model was trained on Base44’s accumulated interaction data rather than general web text, which is what makes it potentially competitive with frontier APIs on this narrow task even without general-purpose reasoning capability.

The move follows a pattern emerging across AI product companies. Rather than relying exclusively on frontier model APIs from OpenAI, Anthropic, or Google, companies with sufficient specialized interaction data are now training purpose-built models to control inference costs and create technical differentiation. The key input is not compute—it is proprietary data from real user sessions at scale.

For product managers, this is a signal worth tracking. Whether Base1 outperforms frontier models on its target tasks is still unproven. But the strategic logic—using accumulated user interaction data as the basis for model differentiation—represents a direction that companies with meaningful usage may increasingly pursue. The era of treating LLM access as a commodity input is narrowing for companies that have the data to do otherwise.