Base44 is rolling out its own large language model, Base1, a year after Wix acquired the vibe-coding startup for $80 million. The Bay Area company says Base1 was trained on a dataset built from tens of millions of real user interactions on its platform, and it is designed to eventually replace frontier models like Anthropic's Opus for app-generation workloads. Base44 crossed $100 million in annual recurring revenue a few months ago, and the model launch is the clearest signal yet that vibe-coding platforms see owning the stack as the only way to stay defensible.
Founder Maor Shlomo framed the decision as a stack-level optimization play rather than a research bet. Owning the model, he said, lets Base44 tune for latency, cost, and efficiency in ways a customer of a third-party API cannot.
“training and owning the model as part of [our] entire stack allows us a lot more optimizations on latency, cost, and efficiency.”— Maor Shlomo, Base44 founder
The competitive backdrop is unforgiving. Swedish rival Lovable, which still runs on external LLMs, said earlier this month it had reached $500 million in ARR — five times Base44's run rate. Cursor, Grok parent xAI, and Claude Code have all pushed into app generation, giving frontier providers their own feedback loops on the same workflows Base44 depends on.
Key facts
- 01Base44 is rolling out Base1, its first in-house LLM, trained on tens of millions of real user interactions from its vibe-coding platform.
- 02Wix acquired Base44 for $80 million one year ago, when the startup was six months old with a team of eight.
- 03Base44 crossed $100M in annual recurring revenue a few months ago; rival Lovable hit $500M ARR earlier this month.
- 04Parent company Wix recently announced a 20% workforce layoff, raising the stakes on Base44's margin profile.
- 05Frontier competitors including Cursor, xAI's Grok, and [Anthropic](/claude)'s Claude Code are pushing deeper into app generation.
Base44 was just six months old and had eight employees when Wix bought it. Since the acquisition, headcount has grown, and the unit is now one of Wix's faster-moving assets at a moment when the parent company recently announced a 20% workforce layoff. In a press release, Base44 said ownership of the model gives it direct control over compute and inference spend and is expected to result in a structurally stronger margin profile over time.
Shlomo expects more applied-AI companies to follow — at least the ones with enough scale and velocity to have enough data. He argues specialization is the wedge against general-purpose frontier models.
Jonathan Userovici, a general partner at Headline whose portfolio includes Mistral AI but not Base44, said data is one of three ingredients of defensibility for AI startups, alongside distribution and tech stack. He described Base44 as fitting a broader pattern of companies leaning into their data and infrastructure as inference costs become a meaningful part of the equation.
Userovici cautioned against underestimating frontier labs, pointing to legal-tech startup Harvey, which abandoned plans to train its own model. But he framed Base44's move in the context of enterprise pressure on inference economics.
Enterprise customers remain a minority of vibe-coding platform users but a growing share of revenue, and cost sensitivity is bleeding into the long tail. Shlomo wants a model more aligned to what Base44 thinks is the right output, more optimized to what users actually accept, and faster and cheaper than calling Opus on every request.
The bet has real risk. Base1 is just rolling out, and the dataset advantage Base44 is leaning on will widen for rivals too — Lovable's user base is generating its own interaction data at a pace that scales with its larger ARR. Frontier labs, meanwhile, can absorb training costs Base44 cannot match, and Claude Code has shown Anthropic is willing to compete directly in the vibe-coding category rather than just supplying it. If Base1 underperforms Opus on the workloads users care about most, the optionality of routing to a frontier model never goes away.
Base44's move is the cleanest test case yet for whether application-layer AI companies can survive by going vertical. Owning distribution through Wix, owning the data through the platform, and now owning the model through Base1 gives Base44 the full stack Userovici described — but it also locks the company into a much larger engineering bill at a moment its parent is cutting costs. If Base1 delivers structurally better margins on $100M+ in ARR, expect every vibe-coding platform with the data to follow within twelve months. If it doesn't, the application layer stays a thin wrapper on whoever ships the best frontier model that quarter.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
