Base44 built its own AI model and the vibe coding arms race just got a lot more expensive

Base44, acquired by Wix for $80 million in 2025, launched its own AI model Base 1 on June 29 to reduce reliance on third-party models and improve margins. The move comes as Base44 reaches $150 million in annual recurring revenue, making LLM inference costs a significant expense. By training on proprietary user data, Base44 aims to offer competitive performance at lower cost, differentiating itself in the crowded vibe coding market.

Base44, the no-code app platform Wix bought for roughly $80 million in 2025, launched a proprietary AI model called Base 1 on June 29, signaling that survival in the vibe coding market may now require owning your own model stack. When Base44 announced Base 1 last week, the press release called it a milestone in the company's evolution from "AI-powered platform to full-stack AI company." That framing is careful and deliberate. What it really means is this: running on third-party foundation models is a cost structure that kills margins at scale, and Base44, now clocking $150 million in annual recurring revenue as of May 2026, has grown large enough to do something about it. The model was trained on what Base44 describes as tens of millions of real user interactions on the platform, which is exactly the kind of proprietary data flywheel that makes a domain-specific LLM defensible. Base44 says Base 1 is already performing competitively against the foundation models it replaces, Claude Sonnet 4 among them, with more improvement expected as training continues. The company hasn't published benchmark comparisons, but the core business case doesn't hinge on the model being better than Anthropic's. It hinges on the model being good enough while costing a fraction of per-token API pricing at scale. At $150 million ARR and accelerating, LLM inference costs become one of the largest line items on the income statement. Every app a user builds, every natural language prompt that gets parsed and executed, every multi-turn conversation with the platform, runs on compute. When that compute is rented from a model provider, the margin math is brutal. Wix's Q4 2025 business solutions revenue, the segment that includes Base44, grew 18% to $153.8 million, but analysts tracking the stock have flagged LLM costs as a real dilution risk. Owning Base 1 is the answer to that specific problem. It's also the answer to a competitive problem that nobody in the vibe coding market has quite solved yet. Cursor is valued at $9.2 billion after its Series B last year. Lovable raised at $180 million. Replit processes over 50 million code executions monthly. They're all building on the same underlying models, paying the same providers, and competing largely on developer experience and integrations. When the model is a commodity bought from Anthropic or Google, differentiation is thin. Base44 just made a move that none of those competitors can easily copy: it built something underneath the product that smaller platforms can't afford and that most haven't accumulated enough proprietary training data to replicate. The timing is also pointed. Enterprise buyers are starting to ask hard questions about uptime guarantees and vendor lock-in tied to third-party model providers. When your app-building platform runs on Claude, you're downstream of Anthropic's infrastructure decisions, pricing changes, and rate limits. Base 1 is Base44's answer to that objection in the sales conversation. What Wix is actually building The Base 1 launch matters beyond Base44 itself because it reflects what Wix is becoming. The company now runs three distinct AI products inside its wider platform: Harmony for mainstream website creation, Vibe for headless AI site building with code access, and Base44 for app creation. The acquisition, struck in June 2025 when Base44 was a 500-day-old startup with Maor Shlomo at the helm, looked opportunistic at $80 million. At $150 million ARR and rising, with a cohort of new user bookings up roughly 46% year over year, it looks like one of the better acquisitions in recent Israeli tech history. Wix has guided for mid-teens revenue growth in 2026, and Base44 is the clearest engine behind that forecast. Base44 reaching $100 million ARR just nine months after the acquisition closed was already a signal that the platform had real product-market fit, not acquisition hype. The $150 million figure from May confirms it. Launching a proprietary model now, when revenue is growing fast enough to justify the investment, is the right sequence. Doing it earlier would have been premature. Waiting would have meant paying API costs on a much larger revenue base for longer. The honest question for Cursor, Replit, and whoever else is competing in this market is whether they can afford to follow. Building and training a general-purpose agent model, one capable of handling multi-turn conversations, tool usage, runtime management, and backend operations, the capabilities Base44 claims for Base 1, is not a six-month side project. It requires proprietary training data at scale, infrastructure investment, and a business large enough to absorb the cost of getting it wrong before it gets it right. Most of the vibe coding field doesn't meet those criteria yet. Base44, backed by Wix's balance sheet and its own substantial recurring revenue, does. Frankly, this is what defensibility looks like in an AI-native business when the underlying models are becoming cheaper and more interchangeable. You stop renting the thing everyone else rents and build something only you can build. Also read: TIDAL becomes the first major streaming platform to strip royalties from fully AI-generated music https://startupfortune.com/tidal-becomes-the-first-major-streaming-platform-to-strip-royalties-from-fully-ai-generated-music/ • Ionic Digital takes Celsius Network's ruins to Nasdaq at a $2 billion valuation https://startupfortune.com/ionic-digital-takes-celsius-networks-ruins-to-nasdaq-at-a-2-billion-valuation/ • A 20-year-old high school dropout just raised $31 million to stop AI data centers from poisoning themselves https://startupfortune.com/a-20-year-old-high-school-dropout-just-raised-31-million-to-stop-ai-data-centers-from-poisoning-themselves/