The AI programming platform Base44, acquired by Wix, began to present the self-study model Base1 to users to support the creation of applications in natural languages. Companies want to use their own models to further optimize their product performance in terms of response speed, reasoning costs and matching results.
Base1 on the line.
The founder of Base44, Maor Shlomo, stated that the placement of model training and product stacks in the same system would help to optimize further in terms of delay, cost and efficiency. According to the company, the first version of Base1 is based on data set development and training generated by “tens of millions of real-user interactions” on the platform.
This means that Base44 no longer relies solely on external base models, but begins to integrate model capabilities into its own product core. As users continue to grow, such data will also accumulate and become an important source of follow-up.
It's not just from the same company.
On the surface, Base44's actions are related to the increased competition in the AI programming track. It was reported that the Swedish start-up company Loveable relied on a large external model and had become a unicorn after financing last year. Shlomo believes that companies that reach a certain scale and accumulate sufficient data will probably train their own models in the future.
However, real pressure does not necessarily come from similar platforms alone. Reports suggest that Frontline AI Laboratories are moving further into the AI programming scene. Cursor and Grok's parent company xAI is now part of the SpaceX system, and Claude Code of Anthropic has become an important player in this field.
Focus on data, distribution and costs
According to Jonathan Userovici, a wind project partner in Headline, AI ' s primary moat comes from three main capacities: data, distribution and technology. Base44 This time, the self-study model was launched, and it was along this path that the model, the user data and the product entry were in their own hands as much as possible.
At the same time, he cautioned against underestimating the competitiveness of the front-line model and, for example, that the legal technology company Harvey had abandoned the self-training model scheme. In his view, the application level AI did not necessarily shift to basic model development on a large scale, but the cost of reasoning had become an increasingly realistic operational variable.
Business clients are also driving this change. Userovici states that clients are not willing to use the most advanced and expensive models in all scenarios and that the market is therefore developing an infrastructure around which models are structured and optimized to maintain similar performance in most use scenarios while controlling costs.
Base44 continues to expand while the parent company is downsized.
Shlomo states that Base44 wants the model to be more closely calibrated to user preferences and to be faster and cheaper in the long term than using external front models. In a press release, the company stated that the control of the model would allow direct management of the arithmetical and reasoning expenditures, with long-term prospects for improving the profit structure.
This is particularly important for the parent company Wix. Wix recently announced a 20 per cent reduction, while Base44 continued to expand after the takeover. The company also announced several months ago that annual recurrent revenues had exceeded $100 million.
However, Base44 ' s current income scale is still lower than Loveable. The latter indicated earlier this month that its annual recurrent income had reached $500 million. Shlomo notes that large-scale engineering inputs around Base1 can help Base44 to position “vertical integration”, i.e. to control distribution, data and infrastructure.