Base44, the vibe-coding platform Wix acquired in 2025, has launched Base1, its own AI model for turning plain-English descriptions into working apps. The rollout, announced on June 29, 2026, makes Base44 one of the first app builders to run on a model it trained itself rather than a general-purpose frontier model.
What This Enables
For anyone building on Base44, Base1 is meant to make the core loop, describing an app and watching it get generated, faster and cheaper. Founder Maor Shlomo says the model is fine-tuned on top of an open-source foundation model and trained on tens of millions of real user interactions from the platform, data no outside lab has. The bet is that a model specialized for one job, generating apps with built-in hosting, auth, and databases, can beat a generic model on latency and cost for that job.
Why It Matters
This is the defensibility move a lot of AI startups are now making. Sitting on top of someone else's model means your margins and your roadmap depend on their pricing and their API. By training Base1, Base44 is trying to own its stack the way search engines eventually built their own infrastructure. For creators choosing a build-with-AI tool, it is a signal to watch whether a specialized in-house model actually produces better apps than a Cursor or Lovable riding a frontier model.
Key Details
Model: Base1, fine-tuned on an open-source base, trained on platform interaction data.
Owner: Wix, which bought Base44 for $80 million in 2025.
Scale: Base44 reports more than 2 million users and roughly $150 million in annual recurring revenue, available through the Wix App Market.
Status: Base1 is rolling out to users; the company frames it as the first step, with the goal of eventually outperforming frontier models on its own workload.
What to Do Next
If you already build on Base44, regenerate an existing app once Base1 reaches your account and compare speed and output quality against your last build. If you are evaluating vibe-coding platforms, treat a proprietary model as one factor, not a guarantee, and test the same prompt across two tools before committing a real project.
