- Published on
Vibe coding platform Base44 launches own model as AI startups seek defensibility
- Authors
- Name
- DeepHorus
Vibe coding platform Base44, acquired by Wix for $80 million just a year ago when it was barely six months old with a team of eight, has begun rolling out its own AI model to support users in creating applications through natural language. This strategic shift comes amid intensifying debates in AI circles about whether frontier models are suitable for all use cases and whether businesses built on third-party models are truly defensible long-term. Base44's move, based in Tel Aviv, addresses both concerns by developing proprietary technology to differentiate itself in a crowded market.
Founded by Maor Shlomo, Base44 aims to eventually outperform frontier models with its custom large language model (LLM). According to Shlomo, “training and owning the model as part of [our] entire stack allows us a lot more optimizations on latency, cost, and efficiency.” This approach could help the platform edge out competitors like Swedish startup Lovable, which achieved unicorn status in its Series A last summer and relies on external LLMs. However, Shlomo expects other major players to follow suit once they achieve sufficient scale and data.
Jonathan Userovici, a general partner at VC firm Headline, identifies data, distribution, and tech stack as the three key ingredients of defensibility for AI startups. Base44 fits this pattern by leveraging a dataset generated from “tens of millions of real user interactions on the platform” to train its first model, Base1. The company plans to continuously expand this dataset, though rivals will similarly grow their own data pools.
The broader competitive landscape includes not only vibe coding startups but also frontier AI labs like Anthropic, whose Claude Code has become a direct player in the space. These labs benefit from access to data and feedback loops that improve their models for app creation. Yet Shlomo believes specialization gives Base44 an advantage, predicting that models will remain general while Base44's tailored approach offers superior performance for specific tasks.
Userovici cautions against underestimating frontier models, citing Harvey, a legal tech startup that abandoned plans to train its own model. He does not expect applied AI companies to become frontier labs en masse but frames Base44's move within the context of rising inference costs. Enterprise customers are increasingly demanding cost-effective solutions, leading to infrastructure optimization and model selection to balance performance and expense.
While enterprise clients still represent a minority of vibe coding platform users, their share of revenue is growing. Cost concerns are prompting users of all sizes to seek more efficient alternatives. Shlomo states, “We want to get a model that is going to be more aligned to what we think is the right thing... faster and cheaper for customers eventually than using the frontier models like Opus.” Base44's ownership of its model positions it to address these demands, though the immediate cost benefits are not yet clear.
Looking ahead, Base44's custom LLM could enhance its defensibility and attract cost-conscious users. The company's ability to continuously refine its model with proprietary data may provide a competitive edge, though it must contend with rapid advancements from frontier labs. As inference costs remain a key driver, Base44's strategy reflects a broader industry trend toward vertical integration and specialization aimed at delivering optimized, cost-efficient AI solutions.
