Base44 Ships Its Own LLM, a Year After Wix's $80M Acquisition

Base44 has put its first proprietary large language model into production, making it what Wix's press room describes as the first app-creation platform to run on a purpose-built LLM. The model — referred to as Base-1 — is live and actively serving users on the platform as of late June 2026.

The timeline matters for context. Wix acquired Base44 in June 2025 for $80 million, framing the deal as an expansion into vibe coding — the shorthand for natural-language-driven software generation that has drawn significant developer and enterprise interest. Within roughly twelve months of that acquisition, Base44's engineering team moved from consuming third-party foundation models to deploying one of their own.

From API Consumer to Model Owner

That shift is worth examining on its own terms. The overwhelming majority of no-code and low-code AI platforms — a category that has expanded rapidly since 2023 — are inference consumers. They wrap GPT-4o, Claude, Gemini, or similar frontier models behind a product layer, differentiating through UX, templates, and workflow tooling. Owning the model layer is a structurally different position: it gives a platform direct control over context handling, fine-tuning objectives, output formatting, and, eventually, inference cost.

For a platform whose core value proposition is generating functional application code from natural-language prompts, those levers are not trivial. App generation puts distinctive pressure on a model — it requires coherent multi-file output, consistent adherence to framework conventions, and tolerance for iterative correction in ways that general-purpose chat or summarization do not. A model trained specifically on that task distribution can, in principle, outperform a generalist frontier model on the narrow benchmark that actually matters to the user.

Whether Base-1 does that in practice is not yet independently verified. What is confirmed is that it is in production, not in a limited beta.

A Platform Moving on Multiple Fronts

Base44's model launch did not arrive in isolation. The platform has been expanding its surface area across several vectors. In March 2026, Base44 launched Superagents, a feature set enabling users to build autonomous AI agents configurable to specific workflows — a capability that sits closer to the agentic orchestration space than to the traditional drag-and-drop app builder category. And at an undated point, the platform launched a presence inside ChatGPT, built on OpenAI's Apps SDK, which positions Base44 as an integration target within what is currently the most widely distributed AI interface.

That OpenAI integration is interesting given the proprietary model announcement. Building on OpenAI's SDK while simultaneously training a competing foundation model suggests Base44 is hedging across distribution channels rather than committing to a single-model monoculture. The ChatGPT integration may bring users in; Base-1 gives the platform the option to serve them on its own infrastructure once acquired.

What Wix Gets From This

From Wix's perspective, the vertical integration story becomes more credible with Base-1 in the picture. Wix has historically been a website-builder competing on ease of use, but the Base44 acquisition was an explicit signal that the company intends to compete in the broader software-creation market — a segment with a much larger addressable opportunity and, for now, less mature tooling.

Running a proprietary model trained on app-generation tasks gives Wix a degree of technical differentiation that is harder to replicate than a UX redesign. It also, over time, creates a data flywheel: every app built on Base44 is a potential training signal that a third-party model provider does not capture.

The risk side of the ledger is real too. Training and maintaining a production LLM is operationally expensive, and the frontier continues to advance at pace. A purpose-built small model optimized for code generation can punch above its parameter count, but it needs to be continuously updated to stay relevant as base capabilities improve across the industry. That is a sustained engineering and infrastructure commitment, not a one-time launch.

Still, the direction is legible. Twelve months from acquisition to a live proprietary model is a fast execution cycle. What Base44 does with that model over the next year — in terms of measurable user outcomes, inference cost reduction, and capability differentiation — will be the more consequential data point.