Base44 trains its own model as vibe-coding startups chase moats
Wix-owned Base44 just launched Base1, its in-house model, to cut its frontier-model bill. The defensibility scramble is about to reshape India's app-building boom too.
The News
Wix-owned Base44 has begun rolling out Base1, its own large language model, to wean its vibe-coding platform off the expensive frontier systems it has leaned on since launch. The move, announced on 29 June 2026, is one of the clearest signals yet that the fastest-growing app-building startups now see owning the model, not just the interface, as the route to a durable business.
Base44 lets people build working software by describing it in plain language. Founder Maor Shlomo says Base1 was trained on tens of millions of real user interactions gathered on the platform, to squeeze out cost and latency that general-purpose models cannot match. "Training and owning the model as part of [our] entire stack allows us a lot more optimizations on latency, cost, and efficiency," he said.
The numbers behind the decision are striking. Wix bought Base44 for 80 million dollars in 2025, when the startup was barely six months old and ran on an eight-person team. By May 2026 the platform had reached 150 million dollars in annual recurring revenue, having crossed the 100 million dollar mark just two months earlier, even as Wix cut 20 percent of its own workforce that month.
Why It Matters
For two years, vibe-coding startups have ridden a simple formula: wrap a slick interface around someone else's frontier model and grow fast. Base44's rival Lovable, which still relies on external models, hit 500 million dollars in annual recurring revenue in June 2026 and became a unicorn on a 200 million dollar Series A last summer. But renting intelligence is not a moat, and the bill scales with every user.
Base44's bet is vertical integration: control the data, the model and the distribution at once. Shlomo is candid that this is not about beating the labs on raw capability, but about building something "faster and cheaper for customers eventually than using the frontier models like Opus." The risk is that the labs come downhill, and they already are, with Cursor aligned to xAI and Claude Code from Anthropic pushing into vibe coding. Owning a smaller, sharper model may be the only way a wrapper startup survives that squeeze.
Indian Angle
India is where this maths bites hardest. The country hosts one of the world's largest developer populations and a deep bench of SaaS firms, from Zoho to Freshworks, many now bolting natural-language app building onto their products. Almost all pay for frontier-model access in dollars, so every prompt carries a rupee cost that swings with the exchange rate and a margin that flows to a foreign lab rather than staying onshore.
Base44's move is therefore a template Indian founders will study. A platform that trains a cheaper in-house model on its own usage data keeps both the margin and the data at home, which matters for the unit economics Indian investors scrutinise after recent funding-winter discipline. It also rhymes with the sovereign-model push under the IndiaAI Mission, where Sarvam and Krutrim are building India-first systems rather than renting frontier capability.
For regulators and enterprise buyers, training on user interactions raises questions the Digital Personal Data Protection framework will have to answer, and Indian banks and insurers weighing these tools will want to know whose model sees their prompts. Defensibility is becoming a compliance argument as much as a commercial one.
FAQ
What is Base1?
Base1 is Base44's own large language model, trained on tens of millions of interactions from its app-building platform. It is designed to handle the platform's specific tasks more cheaply and quickly than general frontier models, rather than to outperform them on every benchmark.
How big is Base44 now?
Base44 reached 150 million dollars in annual recurring revenue by May 2026, up from 100 million dollars two months earlier. Wix acquired it for 80 million dollars in 2025, when it had an eight-person team and was about six months old.
What does this mean for Indian developers?
Indian developers and SaaS firms currently pay for frontier-model access in dollars, exposing them to currency swings and margin leakage. A shift towards cheaper in-house or India-built models, such as those from Sarvam or Krutrim, could lower costs and keep sensitive data onshore.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.