Beyond Frontier Models: Base44 Debuts Custom LLM to Cement Vibe

What happened

Base44, the high-profile vibe coding platform acquired by Wix in 2025, has officially entered a new phase of its technological evolution by launching its own proprietary artificial intelligence model. This strategic move follows a year of rapid growth under the Wix umbrella, which purchased the startup for approximately $80 million when it was only six months old and operated with a team of eight. By rolling out a custom Large Language Model (LLM), Base44 aims to transition from a service built on top of external frontier models to a vertically integrated platform that controls its own infrastructure. This shift is designed to enhance the speed and accuracy of app creation through natural language prompts, often referred to in the industry as "vibe coding." The launch signifies a broader trend among AI-native startups seeking to establish deeper moats against well-funded foundational model providers.

What's new in this update

The centerpiece of this update is the introduction of Base1, the first iteration of Base44’s custom-built LLM. Unlike general-purpose models from labs like OpenAI or Anthropic, Base1 was trained on a highly specialized dataset consisting of tens of millions of real user interactions recorded on the Base44 platform. According to founder Maor Shlomo, owning the entire stack—from the user interface to the underlying model—enables significant performance gains. Shlomo emphasized that this level of integration allows for specific optimizations in latency, cost management, and operational efficiency that are impossible when relying on third-party APIs. The company expects that as its data pool grows with continued platform usage, the model will eventually outperform the more generalized frontier models currently dominating the software development space, providing a more tailored experience for app generation.

Key details

The decision to develop Base1 highlights the intensifying pressure on AI startups to prove their long-term defensibility. Jonathan Userovici, a general partner at the venture capital firm Headline, notes that data, distribution, and the technical stack are the three critical ingredients for a startup's survival in the current AI era. While competitors like the Swedish startup Lovable have achieved unicorn status by leveraging external models, Base44 is betting that platform-specific data will provide a superior edge. This approach also addresses growing concerns from enterprise customers regarding the return on investment for AI tools. As inference costs become a larger portion of corporate budgets, enterprise users are increasingly demanding specialized orchestration and model optimization to prevent costs from skyrocketing while maintaining consistent performance across common app-building use cases. Base44’s move aligns with this demand for cost-effective, specialized performance.

Background and context

Base44’s journey from a niche startup to a cornerstone of Wix’s AI strategy has been remarkably swift. When Wix acquired the company, Base44 had only been in existence for six months, yet it had already captured significant attention for its ability to lower the barrier for software creation. The concept of vibe coding allows users to describe an application’s functionality in natural language, which the AI then translates into functional code and deployment-ready software. However, as the space has matured, the threat from foundational AI providers has loomed larger. Anthropic recently entered the fray with Claude Code, and SpaceX-owned entities like Cursor and xAI’s Grok are increasingly targeting the developer experience. This competitive environment has forced specialized platforms to look beyond user interface design and into the mechanics of model training and specialized data loops to avoid being commoditized by general LLM improvements.

What to watch next

The industry is now watching to see if Base44’s move prompts a wave of model-native shifts among other applied AI companies. While some startups, such as the legal tech firm Harvey, have reportedly scaled back plans to train their own frontier-style models due to complexity and cost, Base44’s focus on a specialized domain may prove more sustainable. Observers will also be tracking whether frontier labs like Anthropic and OpenAI can maintain their generalized performance advantage or if specialized models like Base1 will provide a noticeably better user experience for app development. Furthermore, the success of this rollout will likely influence Wix's broader product roadmap as it integrates Base44's technology deeper into its website-building ecosystem. The competition for data feedback loops remains fierce, and Base44's ability to maintain its lead will depend on the continued growth and quality of its proprietary interaction dataset.