Base44 just launched Base1, its own AI model, after a year of building on Claude, GPT, and other frontier models. Here is what actually changed, and what most headlines are not telling you.
Base1 is a fine-tuned open-source model, not one built from scratch, despite being marketed as proprietary.
What Actually Happened
In late June 2026, Base44 began rolling out Base1, its first in-house AI model, ending a year of building its entire platform on top of external models from OpenAI and Anthropic. The announcement was covered widely, TechCrunch, The New Stack, and several other outlets ran the story within days of each other, which is itself a signal of how closely the AI app-builder space is being watched right now.
Base1 was trained on a dataset built from tens of millions of real user interactions across the Base44 platform, using a reinforcement learning driven training process. Founder Maor Shlomo framed the move as a bid for independence: control over latency, inference cost, and output quality, rather than renting all of that from a third party.
What Base1 Actually Is
Here is the detail most coverage of this story skipped past. Base1 is not a model trained from scratch. Reporting from The New Stack, based on a direct interview with Shlomo, confirms Base1 is a fine-tuned open-source model, not an original foundation model built from the ground up. Shlomo himself acknowledged that the term proprietary model gets used loosely across the industry.
Why this distinction matters
A fine-tuned open-source model is a real, legitimate engineering effort, and it can still deliver genuine cost and speed advantages. But it is a meaningfully different claim than training a new foundation model from scratch, and the marketing language around this launch does not always make that distinction clear.
What Base1 does technically: it functions as a general-purpose agent, handling both conversational back and forth and code generation, including multi-turn requests, tool use, and backend operations, echoing the agentic build tools spreading across other AI platforms right now.
Why Base44 Did This
Cost and latency control. Shlomo has been direct about this: relying entirely on external frontier models means Base44 pays for every inference call and inherits whatever latency those providers deliver. Owning even a fine-tuned model gives the company more room to optimize both.
Competitive defensibility. A growing argument across the AI app-builder space is that companies built entirely on top of someone else’s models are not durably defensible long term. Base44’s bet is that specialization, a model tuned specifically for app-building tasks, can beat a general-purpose frontier model on speed and cost for that narrow job, even if it can never out-reason a frontier model broadly.
A real, growing dataset advantage. Base1 was trained on real interaction data from Base44’s own user base, a dataset that keeps growing as the platform grows, and one competitors cannot simply replicate without the same scale of real usage.
The Real Competitive Context
Two pieces of real context make this launch more interesting than a standalone product announcement.
Annual Recurring Revenue, Mid-2026 (reported)
Base44 passed 150 million dollars in annual recurring revenue in May 2026, just two months after crossing 100 million dollars, genuinely fast growth. But Lovable, its closest direct competitor and the subject of our own Lovable review, reportedly reached 500 million dollars in annual recurring revenue around the same period. Base44 is still growing quickly, but it is not the revenue leader in its own category, and Base1 reads in part as an attempt to close a real, measurable gap rather than simply protect a lead.
The second piece of context is stranger. Base44’s parent company, Wix, announced layoffs affecting 20% of its workforce around the same period Base44 itself was growing headcount and investing in a major new model. Wix runs three AI-powered products, Harmony for website building, Vibe for headless development, and Base44 for full-stack app creation, and Wix’s own data science team reportedly collaborated directly on Base1’s development, suggesting the company is concentrating investment in its highest-growth bet even while cutting elsewhere.
What This Means If You Use Base44
If you are currently building on Base44 or considering it, here is what this launch realistically changes and does not change.
What Might Genuinely Improve
Response speed and cost efficiency are the two things Base44 is explicitly optimizing for with its own model. If Shlomo’s stated goals hold up in practice, you may see faster generation and, potentially, better value from your credit allowance over time, though this review has not independently verified either claim yet.
What This Does Not Change
Nothing about this launch addresses the concerns already documented in our full Base44 review, the platform’s security incident history, its frontend-only export limitation, or its reported reliability issues earlier in 2026. A new underlying model does not retroactively fix infrastructure or export architecture decisions made before it existed.
Should This Change Your Decision?
Not on its own. This is a genuinely significant strategic move for Base44 as a company, and worth knowing about if you are already using or evaluating the platform. But the core trade-offs that should actually drive your decision, security posture, lock-in risk, and realistic pricing, are unchanged by which model happens to be generating your code underneath.
Frequently Asked Questions
What is Base1?
Base44’s first in-house AI model, launched in late June 2026. It is a fine-tuned open-source model, not a foundation model built from scratch, trained on real user interaction data from the Base44 platform.
Is Base1 better than Claude or GPT for Base44 users?
Unclear and unverified independently as of this writing. Base44’s own claims center on speed and cost efficiency for app-building tasks specifically, not general reasoning ability.
Does Base1 fix Base44’s security issues?
No. This launch is unrelated to the separate, previously documented security incidents covered in our full Base44 review.
Why did Base44 build its own model instead of continuing to use Claude or GPT?
Founder Maor Shlomo cited cost, latency, and independence from external vendors as the core motivations, alongside a broader industry argument that app-layer companies built entirely on someone else’s models are less defensible long term.
How does Base44’s revenue compare to Lovable’s?
As of mid-2026, Base44 reported 150 million dollars in annual recurring revenue versus Lovable’s reported 500 million dollars, making Lovable the larger of the two by revenue.
Is this a trained-from-scratch model?
No. The New Stack’s reporting, based on a direct interview with the founder, confirms Base1 is a fine-tuned open-source model, despite being described in some coverage as proprietary.
What data was Base1 trained on?
Tens of millions of real user interactions collected across the Base44 platform, using a reinforcement learning driven training process.
Sources and References
- TechCrunch, Vibe-coding platform Base44 launches own model as AI startups seek defensibility
- The New Stack, Base44 bets a narrow model beats frontier AI for vibe coding
- TheNextWeb, Base44 launches Base1, its own AI model for vibe coding
This article reflects independent research and was not sponsored by Base44, Wix, or any competitor named above. Given the pace of change reported around this platform, confirm current details directly before making a decision based on this coverage.
For the full picture on Base44, including its documented security history, pricing, and reliability record, read our complete Base44 Review. Comparing it against a genuine code-ownership alternative? See our Lovable Review.