Base44, the Wix-owned vibe-coding platform, has begun rolling out Base1, its own custom AI model trained on tens of millions of real user interactions, marking a strategic shift away from relying solely on external frontier models like those from OpenAI and Anthropic. The move reflects a broader trend among AI startups seeking to build defensibility through vertical integration, controlling everything from data collection to model inference in their products.

Base44 was acquired by Wix for $80 million in 2025 when the company was just six months old with a team of eight people. Today, it has grown to 2 million users and reached $150 million in annual recurring revenue as of May 2026, making it one of the fastest-growing platforms in the vibe-coding category, which includes competitors like Lovable and Replit.

Why is Base44 building its own AI model instead of relying on existing ones?

The decision to train Base1 centers on three core business advantages: cost control, latency optimization, and specialization.

"Training and owning the model as part of our entire stack allows us a lot more optimizations on latency, cost, and efficiency," said Maor Shlomo, Base44's founder and CEO.

Maor Shlomo, Founder and CEO at Base44

Rather than paying inference fees to OpenAI or Anthropic every time a user generates code or designs a user interface, Base44 can now run its own model on its own infrastructure, directly controlling compute spending.

Base1 was fine-tuned on top of an open-source foundation model rather than trained from scratch, using data from Base44's own platform. This dataset includes tens of millions of real user interactions, capturing what the AI built, what broke, and what users accepted or rejected. The company also applied reinforcement learning in simulated environments to optimize for both working code and better product decisions.

Shlomo has framed the move as completing Base44's vertical integration strategy, which already includes its own backend and database infrastructure. He also noted that the initiative serves as a hedge against tightening U.S. export controls on frontier models, a concern particularly relevant to Base44's Israeli operations.

How does Base44's approach compare to competitors in the vibe-coding space?

The vibe-coding market is experiencing intense competition from multiple directions. Swedish startup Lovable, which reached unicorn status after its Series A funding round, continues to rely entirely on external large language models and has achieved approximately $500 million in annual recurring revenue, generating about one million new projects per week. Meanwhile, frontier AI labs themselves are entering the space: Anthropic offers Claude Code, xAI (owned by SpaceX) has Grok Build, and Cursor is expanding its coding capabilities.

Shlomo believes that competitors with sufficient scale and data will eventually follow Base44's path. However, he argues that specialization gives Base44 an edge.

"Models are advancing, but they will remain very general in what they can do," said Maor Shlomo.

Maor Shlomo, Founder and CEO at Base44

A narrowly focused model trained specifically on app-building interactions can outperform general-purpose frontier models on that particular task while running faster and cheaper.

What do venture investors say about this strategy?

Jonathan Userovici, a general partner at venture capital firm Headline, identified three pillars of defensibility for AI startups in competitive markets:

• Data: The ability to collect, organize, and learn from proprietary user interactions and feedback loops that competitors cannot easily replicate.
• Distribution: A direct relationship with end users and the ability to reach new customers faster than rivals in the same category.
• Technology Infrastructure: Owning the underlying compute, model, and software stack rather than depending on third-party providers for critical capabilities.

Userovici also noted that enterprise customers increasingly prioritize cost optimization over always using the most advanced model available.

"They don't necessarily see a return on investment when using the latest models for all use cases," said Jonathan Userovici.

Jonathan Userovici, General Partner at Headline

This shift is driving companies to seek systems that intelligently select the right model for each task while keeping inference costs under control.

How to evaluate whether a vibe-coding platform's AI model strategy matters for your team

• Cost Trajectory: Ask vendors whether they own their inference infrastructure or pass through third-party API costs. Platforms with custom models may offer lower per-request pricing as they scale, while those relying on external models face fixed or rising costs tied to frontier model pricing.
• Security and Compliance: Base44 has experienced security vulnerabilities in the past, including a permissions flaw disclosed by Wiz that exposed personally identifiable information and trade secrets. The company now automatically scans every generated app for exposure and misconfiguration issues, a practice worth verifying with any vibe-coding vendor.
• Model Specialization: Evaluate whether a platform's AI model is optimized for your specific use case, such as web app generation or mobile development. A specialized model may produce better results for your workflow than a general-purpose frontier model, even if it has lower overall capability.

Base1 is being introduced gradually and currently sits alongside GPT-5.5 and Claude Opus 4.8 in Base44's model selector, allowing users to choose which model powers their app-building session. Shlomo has stated that the first versions of Base1 are designed to match, not yet exceed, frontier models on app-building quality, with the long-term ambition to eventually outperform them.

The broader trend is clear: as AI startups accumulate sufficient user data and reach meaningful scale, the economics of owning their own models begin to favor vertical integration. Base44's $150 million ARR milestone suggests the company has crossed that threshold, and its investment in Base1 signals confidence that specialization and cost control will become key competitive advantages in the vibe-coding market over the next two to three years.