Base44 Unveils Its Own AI Model to Strengthen Its Position in Vibe Coding

Base44, the vibe-coding platform acquired by Wix for $80 million just a year after its launch—when the startup was only six months old and operated with a team of eight employees—has begun rolling out its own large language model (LLM) to help users build applications using natural language.

The launch comes as debate continues across the AI industry over whether frontier AI models are the best solution for every application and whether companies built entirely on third-party models can remain competitive over the long term. Base44’s latest move reflects both of those discussions.

Although its custom LLM is only beginning its rollout, Base44 believes the model will ultimately outperform general-purpose frontier models for its specific use cases. Founder Maor Shlomo said that “training and owning the model as part of our entire stack allows us a lot more optimisations on latency, cost, and efficiency.”

Initially, the move may appear to be aimed at gaining an advantage over competitors such as the Swedish startup Lovable, which achieved unicorn status during its Series A funding round last summer and continues to rely on external LLM providers. However, Shlomo expects other large players to eventually build their own models as well—particularly companies that have accumulated enough scale, user activity, and proprietary data to justify doing so.

Jonathan Userovici, general partner at venture capital firm Headline, whose portfolio includes AI companies such as Mistral AI but not Base44, believes defensibility for AI startups comes down to three major factors: proprietary data, distribution, and ownership of the technology stack.

That trend is becoming increasingly visible as AI companies with established brands leverage their own datasets and infrastructure to strengthen their competitive position. Base44 follows that approach, saying its first proprietary model, Base1, was developed and trained using a dataset generated from tens of millions of real user interactions collected through its platform.

As the platform continues to grow, so will its dataset. However, competitors are also expanding their own training data. The biggest challenge may not come from other vibe-coding startups but from frontier AI companies moving closer into Base44’s territory. Cursor and xAI, the parent company behind Grok, are now both part of SpaceX, while Anthropic’s Claude Code has also evolved into a significant player in AI-assisted application development.

These developments provide foundational AI companies with valuable user feedback and training data that can further improve their coding models. Despite that, Shlomo believes specialised platforms still maintain a meaningful advantage.

“Models are progressing, but they’ll stay very general in what they can do,” he said.

Userovici, meanwhile, cautioned against underestimating the capabilities of frontier AI providers. He pointed to legal technology startup Harvey, which eventually abandoned its plans to develop its own model. Rather than expecting application-layer AI companies to become frontier labs themselves, he views Base44’s strategy within the broader context of rising inference costs.

According to Userovici, increasing inference expenses have prompted enterprise customers to demand better optimisation. “They don’t necessarily see a return on investment when using the latest models for every use case, so an entire infrastructure is emerging to orchestrate and optimise model selection, ensuring costs stay under control while maintaining similar performance across most workloads,” he explained.

Although enterprise customers still represent a relatively small portion of the overall user base for vibe-coding platforms, they account for an increasing share of revenue. At the same time, organisations of all sizes are becoming more aware of AI operating costs. Developing its own LLM was driven by multiple objectives for Base44, with a lower long-term cost of AI likely among the expected advantages.

“We want to build a model that aligns more closely with what we believe users actually need, one that delivers results people prefer while also becoming faster and more affordable for customers than frontier models like Opus,” Shlomo said.

For Base44 itself, however, financial benefits may not be immediate. The company said in a press release that ownership of its model provides direct control over compute and inference spending, which is expected to create a structurally stronger margin profile over time.

Improved profitability would be welcome news for parent company Wix, which recently announced plans to reduce its workforce by 20%. In contrast, Base44 has continued expanding since being acquired. The company revealed that it surpassed $150 million in annual recurring revenue in May, only two months after crossing the $100 million ARR milestone.

That figure remains below rival Lovable, which announced earlier this month that it had reached $500 million in annual recurring revenue. Nevertheless, Shlomo believes the significant engineering investment behind Base1 will reinforce Base44’s position as the only vertically integrated vibe-coding platform—one that controls its distribution, proprietary data, and infrastructure under a single ecosystem.