Base44 Just Shipped Its Own LLM — What Happens When a No-Code Platform Doesn't Need OpenAI Anymore?

On 29 June 2026, Base44 shipped its own large language model, something the company describes as a first for a no-code app-creation platform. Not a wrapper. Not a router that quietly sends your prompts to Claude behind a nice UI. An actual model. Base1. Trained on Base44's own platform data, with inference delivered through managed infrastructure, and now live in production, serving real users.

The press release called it "a major milestone." That's underselling it. This is a structural shift in how no-code platforms think about their relationship with the AI labs they've been renting intelligence from. And if you're building a business on any platform that depends on OpenAI or Anthropic APIs, you should care about this more than you think.

What Base44 actually built

Let's get the facts straight before we get into what it means. Base1 is not a frontier model built from scratch. That costs billions, and Base44's founder Maor Shlomo is the first to admit it. Instead, Base44 took an open-source foundation model and fine-tuned it specifically for app creation, using data from tens of millions of real user interactions on the platform. The model handles both conversation and code generation, multi-turn interactions, tool usage, and backend operations. Users can now select Base1 alongside Anthropic's Opus 4.8 and Fable 5, plus OpenAI's GPT-5.5 when they prompt the platform.

The thing to understand is the data moat. Base44 can see what people ask for, what the agent builds, what breaks, what gets changed, and whether the app actually works. That last signal — did it work or didn't it — is rare and valuable. Frontier models optimise for everything. Base1 optimises for exactly one thing: building working web apps inside Base44's environment. Shlomo's bet is that specialisation beats generality for this specific job. Shlomo says early versions are faster and cheaper than frontier models like Opus, with the goal of pulling ahead over time.

Why now? Follow the money

Base44 hit $150 million in annual recurring revenue (ARR) in May, just two months after crossing $100 million. Its parent company Wix, meanwhile, just laid off 20% of its workforce. So there's a bit of a split-screen dynamic here: Base44 is the growth engine; the rest of Wix is under pressure. Better margins on Base44's AI inference spend would be extremely welcome at head office.

But the real financial story is upstream. OpenAI's IPO has slipped to 2027 at the earliest, with Sam Altman reportedly refusing anything below a $1 trillion valuation. The company lost $38.5 billion last year. It's burning cash on data centres while ChatGPT's weekly active users have reportedly stalled around 900 million. SoftBank, one of OpenAI's biggest backers, saw its stock fall roughly 11 to 13%, depending on the report, the day the delay rumours surfaced.

If you run a platform routing millions of inference calls through someone else's API, you're paying a tax on every single one. And the company collecting that tax is itself under capital pressure, with an IPO that keeps sliding to the right and losses that would make any CFO sweat. That's not a supplier relationship you want to bet your margin profile on.

The Lovable question

Swedish rival Lovable hit $500 million ARR running entirely on external models. By the revenue numbers, Lovable is winning by a factor of three. But Base44's move exposes something uncomfortable about that lead: if both platforms produce comparable output because they're calling the same underlying APIs, the lead is distribution and brand, not product. Those are fragile moats in a market where UX can be cloned in weeks.

If Base1 starts producing meaningfully better app-creation results at lower cost (and that's still an if, not a given) then Lovable's scale advantage looks more like a timing advantage. Can Lovable keep growing and build its own model before the cost and quality gap becomes real? I wouldn't bet against Lovable. But I also wouldn't want to be the platform still paying per-token to someone else's API when my competitor owns the inference stack. That maths gets uglier every quarter.

Is this part of something bigger?

Base44 isn't doing this in a vacuum. Prime Intellect just raised $130 million at a $1 billion valuation to help enterprises train their own models instead of depending on frontier labs. OpenAI itself built its own inference chip, Jalapeño, with Broadcom. Anthropic is reportedly eyeing custom silicon. DeepSeek is developing its own AI chips. Japan launched Noetra, a sovereign AI consortium backed by 27,500 GPUs and four of the country's largest companies.

The era of "just build on GPT" is ending. Not because GPT isn't good (it is) but because the economics of dependency have become impossible to ignore. Every successful AI application company now faces the same maths: at scale, renting intelligence is a margin tax. Owning it is a competitive advantage. The only question is when you can afford to make the switch.

Base44 answered: now.

What about the other bet?

Not everyone is going the vertical-integration route, and that's worth saying plainly because it's not a sign of weakness. It's a different architectural wager.

Stacker, the no-code platform I edit for, takes the opposite approach, at least as I read it: rather than building its own models, it abstracts model choice behind a governed platform layer. The idea is that no single model is going to be the right answer for every use case, every budget, and every regulatory environment. So instead of picking one and optimising around it, you build a platform that lets organisations route to the right model for the right task, with the right permissions, audit trails, and cost controls baked in.

Both strategies have logic. Base44's approach gives you tighter integration, better unit economics at scale, and a data flywheel that compounds. Stacker's approach gives you flexibility, supplier diversification, and a governance layer that enterprises with compliance requirements actually need. The trade-off is real: vertical integration is more efficient but less flexible; model-agnostic governance is more adaptable but doesn't capture the margin upside of owning the stack.

I suspect we'll see the market bifurcate along these lines. Platforms serving builders who want the fastest, cheapest app creation will vertically integrate. Platforms serving teams that need control, compliance, and model portability will stay agnostic. Both can win. The only losing strategy is pretending the choice doesn't matter.

The takeaway

Base44 shipping its own LLM is not just a product update. It's the first shot in a structural reorganisation of how no-code platforms relate to the AI industry. A year ago, building on someone else's model was the only game in town. Now it's a choice. And increasingly, it's a choice with visible costs.

For no-code builders, the practical implication is straightforward: pay attention to your platform's model strategy. If your entire application stack depends on a platform that depends on a single AI provider, you have two points of failure, not one. If your platform has diversified its model supply chain (or better yet, owns part of it) your risk surface is smaller.

This is only going to accelerate. Base44 describes itself as the first no-code platform to ship its own model. It won't be the last.