Base44 launches its own model 'Base1': the vibe-coding race for defensibility

Base44, the Wix-owned vibe-coding platform, said in June 2026 that it is launching its own AI model called Base1. Base1 was trained on tens of millions of real user interactions on the platform and is currently in a staged rollout. Founder Maor Shlomo said the goal is to reduce reliance on frontier models and directly optimize latency, cost, and efficiency. ASAP summarizes it from TechCrunch's primary reporting.

What Base1 actually is

What Base44 launched is its own AI model, Base1, the core engine that generates vibe-coding output. Base1 was trained on tens of millions of real user interactions accumulated on the Base44 platform, and it is being introduced through a staged rollout rather than an all-at-once switch. The key move is inserting an in-house model directly into a code-generation pipeline that previously relied on external frontier models.

The case Shlomo is making

The rationale for an own model is owning the entire stack to directly optimize latency, cost, and efficiency. Shlomo said that "training and owning the model as part of [our] entire stack allows us a lot more optimizations on latency, cost, and efficiency." He framed the goal as a model that is more aligned and optimized, and eventually faster and cheaper for customers than frontier models like Opus.

The data engine built by a team of eight

Base44 is a vibe-coding startup that Wix acquired for $80 million in June 2025, when it was six months old with a team of eight. Its founder is Maor Shlomo, and the company has said it passed $100 million in annual recurring revenue after the acquisition. A small team rapidly built revenue and user data, and it is now recycling that data as fuel for training its own model.

Why the competitive axis is shifting to defensibility

The real signal here is that the competitive axis for AI startups is shifting from features toward vertical integration and defensibility. Vibe-coding tools can copy one another's features quickly, so a feature-only moat stays shallow. As long as you rent someone else's frontier model, you cannot control your own cost, speed, or character, and you stay exposed to price competition. Owning the model instead creates a loop where user data becomes model quality, and model quality drives more users back in. That is why Base44 is pulling the model itself in as the final piece of its integration.

How to read this move against the landscape

Overlay the neighboring moves and the direction gets sharper. Rival Lovable reached $500 million in ARR in June 2026, and Anysphere, the maker of Cursor, was acquired by SpaceX. The push to own distribution, data, and infrastructure together has become the industry's shared grammar, and Base44's choice pushes that grammar down into the model layer. Still, it is worth flagging that an in-house model being genuinely faster and cheaper than Opus is a stated goal, not a verified result, and that the phrase "staged rollout" itself concedes the risk of an all-at-once switch.

What Korean practitioners should watch

For teams in Korea using no-code and vibe-coding tools, two things follow. First, a platform that switches to its own model can quietly change the quality and character of generated output, so it is safer to keep your own verification standards rather than leaning on the tool alone. Second, once a vendor starts controlling its own costs, prices and policies can shift with little warning. Before locking your output and workflow deeply into one platform, check its portability and export paths in advance; that is the minimum hedge for the user standing on the other side of this defensibility race.

Source: TechCrunch reporting (2026-06-29, "Vibe coding platform Base44 launches own model").