Base44 Launches Its Own AI Model to Strengthen Its Vibe Coding Platform

For most of the past two years, the vibe coding industry has run on borrowed intelligence. Platforms that let anyone describe an app in plain English and watch it get built have almost universally relied on the same handful of frontier models from OpenAI, Anthropic, and Google, paying per token for access to capabilities they did not own and could not differentiate on. Base44, the Wix-owned platform that went from an eight-person startup to a $100 million-plus ARR business in roughly a year, has just made a different bet. The company has begun rolling out Base1, its own proprietary AI model trained specifically for building applications, in a move that founder Maor Shlomo hopes will eventually let it outperform the general-purpose frontier models the rest of the industry still rents.

This is not a small decision, and it is not happening in isolation. It reflects a much larger question that is now circulating across the entire applied AI industry: if every company is building on top of the same handful of foundation models, what actually makes one AI product more valuable or more durable than another? Base44's answer is to stop renting and start owning. Here is what that actually means in practice, why the company is making this bet now, and what the broader vibe coding landscape looks like as a result.

What Base44 Actually Announced

The company began rolling out its first proprietary model this week, marking a significant shift from its previous approach of building entirely on top of third-party frontier models. Base44, the vibe coding platform that Wix acquired for $80 million just one year ago, when the company was barely six months old and had a team of eight, has started rolling out its own AI model to support its users in creating apps with natural language. The model is called Base1, and it is being introduced gradually to the platform's user base rather than as an immediate, complete replacement for the third-party models Base44 has relied on until now.

Importantly, this is not a model trained from scratch in the way a frontier lab like OpenAI or Anthropic builds its flagship systems. Base44, an AI-powered app-building platform, has launched its first proprietary AI model, dubbed Base One. It is a fine-tuned open-source LLM trained specifically for vibe coding web applications. Shlomo has been transparent about this distinction in interviews. "Base1 is built on an existing open-source model. Building a frontier model from scratch requires several billion dollars. Open-source models are already incredibly powerful and are giving the largest AI labs real competition. We took one of those models and optimized it specifically for Base44's use case, building applications. That's our bet: a specialized model inside a platform will outperform a generic model while also being faster and more efficient."

"Training and owning the model as part of our entire stack allows us a lot more optimizations."
- Maor Shlomo, Founder and CEO, Base44

How Base1 Was Actually Trained

The training data behind Base1 is one of the more interesting details of this launch, because it is not synthetic or licensed data. It comes directly from the platform's own usage. The company says the first iteration of its LLM, Base1, was developed and trained on a dataset generated from "tens of millions of real user interactions on the platform." That figure represents a meaningful proprietary asset that competitors building on top of general-purpose frontier models simply do not have access to in the same form: a record of exactly how real, non-technical users describe the apps they want to build, and what kinds of generated code actually satisfied their intent versus what required correction.

The technical training methodology relies on reinforcement learning grounded directly in platform tasks rather than abstract benchmark performance. The training approach is reinforcement learning-driven. Base44 runs the model repeatedly against real platform tasks, building and editing applications, scoring outputs as good or bad, and feeding that signal back to update the model's weights, he says. This approach means Base1 is being optimized specifically against the kinds of tasks Base44 users actually attempt, rather than against the broad, general-purpose benchmarks that frontier labs use to evaluate their models across a much wider range of use cases.

Notably, this was not a project Base44 undertook entirely on its own. The model development was a collaboration between Base44 and Wix's existing machine learning and data science team, which previously built Wix Harmony. That collaboration gave Base44 access to machine learning expertise and infrastructure that a standalone startup of its size, even one generating over $100 million in annual recurring revenue, would likely have struggled to assemble independently and quickly.

How the Model Fits Into Base44's Existing Workflow

Base1 is not designed to function as a standalone, general-purpose chatbot. It is built specifically to operate within Base44's existing application-building system. The model itself is tuned to Base44's agentic harness, which includes its tooling, instructions, and the agent's operation within the platform. That tight coupling between the model and the surrounding agentic infrastructure, the tools it can call, the instructions that govern its behavior, the way it operates within a multi-step build process, is part of what Base44 is betting will give it a performance edge that a more generic frontier model, however capable in the abstract, would struggle to replicate without similar fine-tuning to the platform's specific operational context.

A Genuine First in the Vibe Coding Category

Base44 and several outlets covering the launch have framed it as an industry first within the specific category of consumer-facing app-creation platforms. The Wix-owned vibe coding platform becomes the first app-creation tool to deploy its own large language model in production. That distinction matters because nearly every comparable platform in this space, despite raising significant funding and reaching substantial revenue, has continued to operate purely as an orchestration layer on top of externally sourced models.

The competitive landscape makes this first-mover claim more notable. Platforms like Replit, Bolt, and Lovable are all competing in the AI-assisted development space, but none have deployed a proprietary model in production. Base44 claiming that first-mover position gives it a potential structural advantage, particularly on cost. The New Stack's reporting adds an important caveat to the first-mover framing that is worth taking seriously. The company is claiming it as an industry first: the first app-creation platform to launch its own model. The claim carries an asterisk, though. Base44 founder and CEO Maor Shlomo confirms in an interview with The New Stack that Base One is a fine-tune on an existing open-source base, not a model trained from scratch. The term "proprietary model" gets used loosely across the industry. That distinction is worth keeping in mind: Base1 represents genuine technical investment and a real strategic shift, but it is built on an open-source foundation rather than being an entirely novel architecture developed from the ground up.

Why Building a Proprietary Model Makes Strategic Sense Now

The decision to invest in a custom model is rooted in a broader, increasingly urgent debate playing out across the applied AI industry about long-term defensibility. The move comes as the discussion in AI circles has intensified over whether frontier models are best suited for all use cases. A related question is whether businesses built on top of someone else's models are truly defensible long-term.

Cost has become a central pressure point driving this kind of decision across the industry, not just for Base44. Inference costs have become a meaningful part of the equation. That cost pressure has driven change that enterprise customers are now demanding. "They don't necessarily see a return on investment when using the latest models for all use cases, so an entire infrastructure is being set up to do orchestration and optimization to select the right models for them so that costs don't skyrocket while maintaining the same or similar performance across the majority of use cases."

Shlomo has framed the underlying logic behind owning the model in terms of what it unlocks across the entire technology stack, not just raw model performance. "Training and owning the model as part of our entire stack allows us a lot more optimizations," Shlomo said. The company believes that controlling the model layer directly, rather than treating it as an external dependency accessed through an API, allows for tighter integration with the rest of Base44's infrastructure in ways that compound advantages across latency, cost, and the specificity of generated output.

• Lower per-request inference costs compared to paying frontier lab API pricing at scale
• Reduced latency from a model fine-tuned specifically for the platform's narrow task domain
• Tighter integration with Base44's existing agentic tooling and harness
• A growing proprietary training dataset that competitors using only third-party models cannot replicate
• Greater long-term control over the technology stack as enterprise customers demand cost predictability

The Vertical Integration Bet

Shlomo's stated ambition extends well beyond simply launching a single model. The framing he has used publicly positions Base1 as the foundation for a much broader strategic positioning within the vibe coding category. Shlomo is betting that the "huge engineering effort" to develop Base1 will cement Base44's positioning as the "only vertically integrated vibe-coding application," meaning, in Userovici's terms, a player that owns its distribution, data, and infrastructure all at once.

This vertical integration framing is a deliberate point of differentiation against Base44's most direct competitor in the consumer vibe coding space. Base44's vertical integration strategy could help it differentiate from competitors like Swedish startup Lovable, which reached unicorn status in its Series A round and still relies on external LLMs. Owning the distribution channel, the proprietary user interaction data, and now the model itself gives Base44 a structurally different position than a platform that has scaled rapidly while remaining entirely dependent on whichever frontier lab offers the best price-performance ratio at any given moment.

The Real Threat May Not Be Other Vibe Coding Startups

Perhaps the most interesting strategic insight from this launch is Shlomo's own assessment of where the most serious long-term competitive threat is actually coming from, and it is not the obvious answer. While Lovable, Replit, and Bolt are the companies most directly compared to Base44 in coverage of the vibe coding space, the company's leadership appears more concerned about a different category of competitor entirely.

The bigger competition may not be vibe-coding startups at all but instead come from frontier AI labs that are getting closer to Base44's home turf. Cursor and Grok's parent company xAI now both belong to SpaceX, and Claude Code has become a vibe coding player in its own right. This gives Anthropic and other foundational AI providers access to data and feedback loops they can use to improve models for app creation.

This is a genuinely consequential dynamic. Anthropic's Claude Code, originally positioned as a developer productivity tool for professional engineers, has steadily expanded into capabilities that overlap meaningfully with what consumer vibe coding platforms offer. The same is true of Cursor and the broader coding agent ecosystem that companies like xAI are investing in heavily. These frontier labs have access to vastly more computational resources, larger research teams, and broader user feedback loops than any applied AI startup, including one with Base44's revenue trajectory, could hope to match directly.

Shlomo's argument is that specialization, rather than scale, is what gives Base44 a defensible position against this broader category of well-resourced rivals. Shlomo argues that specialization gives Base44 an edge. "Models are progressing, but they'll stay very general in what they can do," he predicted. The bet here is essentially that a model deeply optimized for one narrow task, building applications from natural language descriptions on Base44's specific platform, can outperform even a vastly more capable general-purpose model on that specific task, similar to how specialized tools often outperform generalist tools within their narrow domain even when the generalist has more raw capability overall.

A Skeptical Counterpoint

Not everyone covering this strategic shift is convinced specialization is a durable advantage, and it is worth presenting that skepticism directly rather than treating Base44's framing as the final word. One cited expert pointed to a cautionary example from a different applied AI vertical entirely.

Userovici cautions against underestimating frontier models, pointing to the example of legal tech startup Harvey, which abandoned plans to train its own model. He does not expect applied AI companies to become frontier labs en masse, but sees Base44's move as part of a broader trend toward cost optimization.

Harvey's experience is instructive precisely because it followed a similar logic to what Base44 is now pursuing: a vertical-specific applied AI company with substantial funding and a clear, narrow use case decided that owning a proprietary model would provide a defensible advantage, only to ultimately conclude that the pace of frontier model improvement made that bet less attractive than continuing to build on top of general-purpose models and focusing differentiation efforts elsewhere in the product. Whether Base44's situation is different enough, given its specific combination of proprietary interaction data, vertical integration with Wix's infrastructure, and a narrower, more constrained task domain than legal work, to avoid a similar outcome remains an open question.

The Broader Base44 Growth Story

To understand why this bet matters, it helps to understand just how quickly Base44 has grown since its founding, a trajectory that several outlets have described in genuinely remarkable terms. Shlomo founded Base44 in late 2024, shortly after completing an extended period of reserve duty following the October 7 attacks. The startup was among the first companies to recognize the potential of large language models to enable non-programmers to build software simply by describing what they wanted in natural language, even before "vibe coding" became one of the defining trends of the AI industry. Within months, the company had attracted more than 100,000 users and signed partnerships with companies including eToro and Similarweb.

The acquisition by Wix happened remarkably fast, and at a price that now looks notably favorable given the platform's subsequent growth. Just six months after it was founded, with only six employees and no outside funding, Base44 was acquired by Wix for at least $80 million. Last November, Wix reported that Base44 had surpassed 2 million users. The platform was also adding more than 1,000 paying subscribers a day and had reached an annual recurring revenue run rate of $50 million.

That growth rate has continued to accelerate since the acquisition. By May, that figure had climbed to $150 million. According to the company, Base44 is now the largest AI-powered application creation platform in North America.

"We're just getting started here in this lane of training a model."
- Maor Shlomo, on Base1 being the first in a planned series of model releases

Base44's Recent Feature History Shows a Pattern of Aggressive Shipping

The Base1 launch fits into a broader pattern of rapid, consistent product investment that has characterized Base44 since the Wix acquisition rather than being an isolated, one-off strategic pivot. Since the acquisition, Base44 has kept shipping aggressively. In March 2026, the platform launched a ChatGPT integration alongside "Superagents," a feature enabling autonomous AI agents to handle complex multi-step tasks within user-built applications.

Importantly, the launch of a proprietary model does not mean Base44 is abandoning access to third-party frontier models for users who prefer them. The platform also supports third-party models like GPT-5, giving users flexibility to choose which AI backbone powers their projects. The platform's own marketing materials reinforce this multi-model approach as a core part of the product. Get access to the latest AI models as they launch. Base44 automatically selects the best model for your project, or you can choose the one that fits your build, your style, and your workflow. Base1 is being introduced as a new option within that existing multi-model framework, with the longer-term aspiration that it eventually becomes the default and preferred choice for most app-building tasks on the platform, rather than as a forced, exclusive replacement for every model Base44 has previously supported.

What Comes Next for Base1

Shlomo has been clear that this initial release represents the beginning of a longer-term model development effort rather than a completed initiative. Base One is the first in what Shlomo says will be a series of releases, with larger models and deeper product integration planned. "We're just getting started here in this lane of training a model," he says.

At the same time, the company's own leadership has been measured about expectations for how dramatically this single release will reshape the competitive landscape in the near term. Shlomo says he doesn't expect the new model's launch to fundamentally shift the competitive landscape, but sees it as a quality and efficiency advantage as Base44 iterates on future versions. That framing is worth taking seriously: Base1 is best understood as the opening move in a longer-term strategic commitment to model ownership, not a single launch event expected to immediately settle the question of whether vertical integration beats reliance on frontier models in the vibe coding category.

Growth Has Come With Real Security Scrutiny

It is worth noting that Base44's rapid scale has not come without challenges, particularly around platform security, since the company's approach to handling that scrutiny is relevant context for understanding how seriously it is treating engineering investment more broadly, including in Base1. The platform's rapid growth has also exposed it to the security challenges facing the broader vibe-coding industry. Among other incidents, Wiz disclosed a serious permissions flaw that affected applications built on the platform and exposed personally identifiable information and trade secrets belonging to thousands of organizations. Cybersecurity company Imperva also identified several critical vulnerabilities that could have allowed attackers to access sensitive information and take control of applications.

Shlomo has described the company's response to these incidents as a significant ongoing investment area. "Today, our platform includes multiple layers of security," he said. "Every application is scanned automatically for issues such as data exposure, insecure code, and configuration mistakes. Organizations can also define strict permissions and control who can use each application." The fact that Base44 is simultaneously investing in proprietary model development and in addressing real, publicly disclosed security vulnerabilities suggests a company attempting to mature its engineering practices across multiple dimensions at once as it scales toward a much larger enterprise customer base, a transition that brings significantly higher stakes than its earlier growth phase serving primarily individual creators and small businesses.

What This Means for the Vibe Coding Industry Going Forward

Base44's decision to build Base1 is likely to be watched closely by every other vibe coding platform currently operating purely as an orchestration layer on top of rented frontier models. If Base1 demonstrates measurable cost, latency, or output quality advantages over the next several model iterations, it would provide a concrete proof point for the broader industry argument that vertical integration, owning your own fine-tuned model rather than remaining permanently dependent on external providers, is a viable path toward genuine defensibility in a market where the underlying foundation models are increasingly commoditized and accessible to everyone.

If, on the other hand, Base1 struggles to keep pace with the rapid improvement curve of frontier models from labs investing billions of dollars and vast research teams into general capability gains, Base44's bet could end up resembling Harvey's earlier retreat from proprietary model training: a reasonable strategic hypothesis that ultimately proved less efficient than simply continuing to rent the best available model and differentiating through product experience, distribution, and proprietary data instead. Either outcome will offer a meaningful data point for the dozens of other applied AI companies currently weighing the same build-versus-rent decision Base44 has just made publicly and decisively.

Related Topics: #Base44 #VibeCoding #Wix #AIModel #ArtificialIntelligence #Lovable #ClaudeCode #AppDevelopment #Technology #StartupNews