# Base44 launches Base1, its own AI model for vibe coding > Source: > Published: 2026-07-01 12:46:43+00:00 *Base44 spent its first year building apps on other companies’ AI models. Now it has built its own. It is betting that owning the model is the only way to survive the vibe-coding gold rush.* The Tel Aviv startup lets anyone build a working app by describing it in plain language. It has now launched its first proprietary AI model, called Base1. Base44 says the model is [already in production](https://www.globenewswire.com/news-release/2026/06/29/3319000/0/en/Base44-Becomes-First-App-Creation-Platform-to-Launch-Its-Own-Proprietary-LLM-Base-1-Marking-a-Major-Milestone-in-the-Company-s-Technology-Vision.html) and serving users. It claims a first here. No other app-creation platform, it says, has shipped its own model rather than renting one from a frontier lab. Base44 trained Base1 on a dataset drawn from tens of millions of real user interactions on its platform. That data is the point. Every app someone builds teaches the model what a good result looks like. It is a feedback loop the startup owns and its rivals do not. Base1 is a general-purpose agent, built to both hold a conversation and write code. It handles multi-turn requests, tool use, and backend operations. That echoes the [agentic build tools](https://thenextweb.com/news/roblox-ai-assistant-agentic-tools-planning-procedural-models) spreading across other creative platforms. Founder and chief executive Maor Shlomo framed the launch as independence. “Having our own model means we can continue to improve it over time,” he said. It also makes Base44 less dependent on external vendors. He was blunt about the difficulty, calling it “not a trivial effort” that needs deep expertise, heavy infrastructure, and rich data most companies lack. ## Why a vibe-coding app builds its own model Base44’s rise is one of the fastest in recent tech. Shlomo founded it in 2024. Six months later, with a team of eight, he sold it to website giant Wix for $80mn. The platform has grown quickly since. It passed $150mn in annualised revenue in May, two months after crossing $100mn. Building a model is expensive and hard, so why do it? The honest answer is money and control. Renting a frontier model from OpenAI or Anthropic means paying for every token a user’s app generates. The model should be “faster and cheaper for customers eventually than using the frontier models like Opus,” Shlomo told [TechCrunch](https://techcrunch.com/2026/06/29/vibe-coding-platform-base44-launches-own-model-as-ai-startups-seek-defensibility/). For the company, owning the stack means direct control over compute and inference spend. That, it says, should produce a “structurally stronger margin profile over time.” That margin maths matters more than usual right now. Parent company Wix recently [cut 20 per cent of its workforce](https://thenextweb.com/news/wix-is-cutting-20-of-its-workforce-as-a-strong-shekel-and-ai-competition-squeeze-the-website-builder-from-both-sides), squeezed by AI competition and a strong shekel. Better margins from Base44 would be welcome news. Base44, by contrast, has been hiring. ## The defensibility question The launch lands in the middle of an argument raging across AI. If your whole business runs on someone else’s model, is it really defensible? A wrapper around GPT or Claude can be copied, and the lab underneath can always move up the stack. Base44 now joins a wave of companies [bringing model development in-house](https://thenextweb.com/news/meta-muse-spark-msl-first-model). That threat is not hypothetical. Anthropic’s Claude Code has become a serious vibe-coding tool in its own right, feeding the lab data on how people build apps. Cursor and Grok’s parent xAI now both sit inside SpaceX. The frontier labs are moving onto Base44’s turf, and they arrive with far deeper pockets. Design and build platforms like [Figma](https://thenextweb.com/news/figma-config-code-layers-ai-skills-plugins-animations) keep racing to embed AI of their own. Shlomo’s bet is specialisation. He argues frontier models “will stay very general,” while a model tuned on app-building data can beat them at that one job. He frames Base44 as the “only vertically integrated” vibe-coding platform. It owns its data, its distribution, and its infrastructure at once. Jonathan Userovici, a general partner at VC firm Headline, calls those three ingredients the recipe for defensibility. ## The case for caution Not everyone is convinced. Userovici also warned against underestimating frontier models. He pointed to legal-AI startup Harvey, which abandoned its own plans to train a model. Frontier labs keep improving, and a general model that is good enough can undercut a specialised one that took a huge effort to build. There is also the plain matter of scale. Base44’s $150mn in revenue is real. But Swedish rival Lovable, which still runs on external models, says it has hit $500mn. Base1 is a first version, and Base44 admits it is only “already demonstrating competitive performance” against the models it replaces, not beating them yet. The company is betting that its data advantage compounds faster than its rivals close the gap. The wider signal is what makes this interesting. A year ago, “just build on GPT” was a viable strategy. Now the successful AI application companies face pressure to own their models, their data, and their costs. The [price of running AI](https://thenextweb.com/news/microsoft-claude-code-retreat-ai-cost), and the reach of the labs, have made the rented approach look fragile. Base44 is the first vibe-coding platform to make that leap. It will not be the last. ## Get the TNW newsletter Get the most important tech news in your inbox each week.