Base44 Launches Custom AI Model to Cut Costs, Boost Defensibility

Base44 builds its own model

Base44, the vibe coding platform Wix acquired for $80 million in 2025, has begun deploying its own large language model to power app creation through natural language. The move signals a broader shift among AI application companies wrestling with two critical questions: whether frontier models suit every use case, and whether businesses built atop third-party AI can sustain competitive advantages.

The Bay Area startup developed Base1 using a dataset drawn from tens of millions of real user interactions on its platform. Founder Maor Shlomo said owning the full stack—from model training to deployment—enables "a lot more optimizations on latency, cost, and efficiency" compared to relying on external providers.

Base44 now claims more than $100 million in annual recurring revenue, though that trails competitor Lovable, which reported $500 million in ARR earlier this year and achieved unicorn status during its Series A. Unlike Lovable, which uses external LLMs, Base44 is betting that vertical integration will prove decisive as the market matures.

Why it matters

The economics of inference are forcing AI application companies to rethink their infrastructure. Enterprise customers increasingly demand cost optimization alongside performance, creating pressure to move beyond one-size-fits-all frontier models. Base44's approach—controlling distribution, proprietary data, and now its own model—represents one path to defensibility as competition intensifies from both specialized startups and foundational AI labs expanding into adjacent markets.

The defensibility calculus

Jonathan Userovici, a general partner at Headline whose portfolio includes Mistral AI, identified three pillars of AI startup defensibility: data, distribution, and technology stack. Base44's strategy addresses all three simultaneously, though Userovici cautioned that the path isn't universal. Legal tech startup Harvey, for instance, abandoned plans to train its own model.

Shlomo expects other players with sufficient scale and data velocity to follow Base44's lead. Yet the competitive landscape extends beyond vibe coding rivals. Anthropic's Claude Code has entered the space directly, while Cursor now shares corporate ownership with xAI through SpaceX, giving frontier labs access to feedback loops that can improve their own app creation capabilities.

Shlomo believes specialization provides an edge. "Models are progressing, but they'll stay very general in what they can do," he said, arguing that Base44's focus on a specific use case will outweigh the resource advantages of larger AI providers.

Cost pressure drives change

Inference expenses have become a meaningful line item for AI companies, particularly as enterprise adoption grows. Userovici noted that customers "don't necessarily see a [return on investment] when using the latest models for all use cases," spurring demand for orchestration systems that route queries to appropriate models based on cost-performance tradeoffs.

For Base44, the calculus involves both customer pricing and internal margins. The company stated that model ownership provides "direct control over compute and inference spend, expected to result in a structurally stronger margin profile over time." That timeline matters for parent company Wix, which recently cut 20% of its workforce. Base44, by contrast, has expanded headcount since the acquisition.

Shlomo characterized the model development as a "huge engineering effort" aimed at positioning Base44 as the only vertically integrated vibe coding platform—a company controlling its entire value chain from user acquisition through model inference.

These details were first reported by TechCrunch.