Talks of millions from vibecoding are everywhere: a million in 17 days, a million by October, a million from one prompt. I decided to investigate: how many projects have truly crossed the threshold of $100,000 in monthly revenue or $1,000,000 in total income (including exit value). The picture turned out to be much more modest than expected. If we exclude projects whose product itself is an AI solution, very few startups based specifically on vibecoding remain.
In this analysis, I do not conflate ARR (Annual Recurring Revenue) and actual revenue. ARR is a projection of the current month multiplied by twelve, not money earned. Where a source indicates ARR, I note it and recalculate the monthly revenue separately. The order of projects in the table below is determined not by revenue size, but by the reliability of financial data confirmation.
Projects that have crossed the $100,000/month or $1 million total income threshold
The list includes projects with public reports of revenue from $100,000 per month or cumulative income from $1,000,000.
-
Medvi: $401 million in the first year
Telemedicine specializing in GLP-1 weight loss drugs. Founder Matthew Gallagher launched the project in September 2024 with an investment of about $20,000. According to publications, the first year brought $401 million in revenue, about 250,000 clients, and $65 million in profit with a 16% margin. Staff: two people (Gallagher himself and his brother Elliot), the rest are contractors.
- Vibecoding: Code, website, and agents were written using Claude, Grok, and ChatGPT. Visuals for advertising — Midjourney and Runway. Client calls — ElevenLabs.
- Business essence: The main part of the business — doctors, pharmacy, prescription writing, and legal framework — is outsourced through CareValidate, OpenLoop Health, and LegalZoom. Gallagher focused on assembly and traffic.
-
Base44: sold for $80 million
An application builder from text description, created by Maor Shlomo. Launched in early 2025. In May, the company showed $189,000 in monthly profit. In June, Wix announced its acquisition for $80 million in cash. At the time of the deal, the project had 250,000 users. Shlomo was the sole owner, but at the time of sale, eight people worked for the company, who received $25 million from the deal amount as retention.
- Vibecoding: Shlomo deliberately built the codebase so that most of it was generated by models, and used prompt routing between different models. This is one of the purest examples where the product itself is vibecoding and built using the same method.
- Confirmation: Financial performance is confirmed by the deal document, not a tweet.
-
Peec AI: $4M ARR (up to $833,000/month)
A Berlin-based company that tracks how a brand is represented in AI search (ChatGPT, Perplexity, Gemini). Founder Marius Meiners. The first prototype was built on v0 from Vercel in about a day and a half. The product was then developed in a traditional way by a team. In November 2025, the company announced a Series A of $21 million, with a total of $29 million raised. At that time, Peec AI had 1300+ connected companies and ARR of about $4 million (approximately $333,000 per month). Later, ARR grew to $10 million (about $833,000 per month).
- Vibecoding: Used to quickly create a prototype and test demand, but the main product was built traditionally.
-
Wave AI: $7M ARR
An application for recording, transcribing, and retelling audio, focused on casual conversations, not meetings. Author Josh Moorer, former Uber GM for New York, not a programmer. He built the first working prototype in Swift in a day using ChatGPT. The project reached $7 million ARR with $3 million in annual profit and about 4000 paid subscribers.
- Vibecoding: Prototype created with AI.
- Feature: High costs for accessing the best models to ensure transcription quality, which is the key product.
Projects that did not reach the threshold, but are often mentioned
These projects often appear in selections, but their financial performance does not meet the specified threshold.
-
Cal AI: $2M monthly revenue, but the label is questionable
An application for counting calories from a photo of food. Authors Zach Yadegari and Henry Langmack. Launched in May 2024. By July 2025 – 8.3 million downloads and, according to CNBC Make It, about $1.4 million in gross monthly profit, and $2 million in monthly revenue is also mentioned. It passes in terms of money, but not in terms of the vibecoding label. TechCrunch revealed that Yadegari wrote the code himself, starting programming at the age of ten. Anthropic and OpenAI models are only used for image recognition within the product.
-
fly.pieter.com: $1M ARR, but peak $87,000/month
A browser-based flight simulator by Pieter Levels. Basic prototype in 30 minutes. $1M ARR was announced in February 2025, 17 days after release. Peak monthly revenue was around $87,000. Monetization: paid aircraft models and items for $29.99. Levels’ method: «one prompt — one layer.» $1M ARR is an annual projection, not a million earned. The project did not reach the $100,000 monthly revenue threshold.
-
Stanley: $1M ARR, but not revenue
An AI content tool created by Vitaly Dodonov and Jon Hu in 14 days. The LinkedIn version was released in June 2025 and showed $200,000 ARR after six weeks. The team passed the $1M ARR mark by October 2025. «14-day app» refers to the first working cut, not the full product. The main value was added after launch. Again, this is ARR, not actual revenue.
-
Lumoo: €700,000 ARR, about €58,000/month
A platform for fashion and retail brands by Henrik Skagerlind Fast and Peter Törngren. Offers photorealistic visualizations and virtual try-on. According to the Lovable case, in nine months the product reached €700,000 ARR, connecting more than fifteen brands. The goal is €1M ARR by November 2025. Monthly revenue is about €58,000, which is below the threshold. The project’s success is due to the founders’ industry connections, not a generated interface.
-
Plinq: $456,000 ARR, about $38,000/month
A background and safety check app by Sabine Matos. Built entirely on Lovable. 6,000 ARR was reported, R3 million raised, and over 10,000 users. 6,000 ARR is about $500 a month, and that’s in the best interpretation. The case is interesting because the app gathered an audience before it made money. Investments came after users appeared, not because the first version was generated by a tool.
General Conclusions and Key Success Factors
Analysis shows that project success is not limited to cheap, AI-generated code. There are several common traits among successful startups.
Team Size at Launch
Four out of five successful cases started with one or two people. This is not so much a credit to vibecoding as it is an indicator that a product can be built faster than a team can be formed.
Infrastructure on Others’ Shoulders
Successful projects actively used third-party services and APIs:
- Medvi outsourced doctors, pharmacies, and lawyers.
- Wave AI used third-party APIs for transcription.
- Cal AI — for recognition and product database.
- Lumoo integrated into existing brand workflows.
Only what was not on the market had to be built from scratch.
Money doesn’t come from cheap code
No product won solely due to cheap or generated code. Money came from other factors:
- Cal AI — through influencers.
- Stanley — through distribution via LinkedIn.
- Medvi — through contractors with licenses.
- Wave AI — through expensive inference (model computations).
- Plinq — through an audience that appeared before revenue.
Code generation shortens the path to launch but is not the main source of income.
Multimodel Verification Method
Almost all teams in the selection use the trick: run one task through several models and compare the results. This is how Peec AI checked demand, Base44 routed prompts, and Levels assembled game layers, changing the model for each task. This allows for quick hypothesis testing and finding optimal solutions. You can test this approach in an evening: take your current prompt and run it through three different models to see the discrepancies and choose the best option.
Frequently Asked Questions
What CPM should be considered profitable for hypothesis testing?
For hypothesis testing, not only CPM is important, but also the speed of data acquisition. If AI tools allow for quick creative generation and campaign launches, then even with a relatively high CPM, hypothesis testing is cheap due to reduced time and labor costs. A profitable CPM is one that provides sufficient data volume for decision-making with minimal launch costs. For example, if the price per one finished clip, generated by AI, allows launching 100 variants for the same budget as 10 variants from a human, it is profitable.
What’s cheaper: banner advertising or bloggers?
A direct price comparison is difficult, as the result is different. Banner advertising (how much does 1 million impressions via banner cost) can be cheaper in terms of CPM, for example, CPM cheaper than 50 rubles per 1000 impressions. But bloggers provide native integration and audience trust. For a budget alternative to blogger seeding, you can consider a mass-posting package price which will be lower, but the reach may be less targeted. Cheaper than influencers can only be your own traffic or very targeted advertising.
Is it worth investing in AI for content creation?
If the goal is to reduce content production costs and increase speed, then is AI worth it — definitely yes. For example, the cost of generating 100 AI videos will be many times lower than creating them manually. The cost of uniqueizing one video with AI also significantly reduces costs. However, it is important to remember that price vs result is a key factor. AI speeds up the process, but a quality result still requires human control and refinement.
How to estimate the cost of AI video production?
The price per minute of AI video with an avatar can vary. It is important to consider not only generation, but also how much it costs to cut 1 hour of stream into short clips using AI. This allows for a lot of content from one source. The cost of publishing 500 videos, if they are generated or cut by AI, will be significantly lower than manual work. What is cheaper: banner or target — this is a question of effectiveness, not just price, and AI can increase the effectiveness of any channel through personalization and scale.
Conclusion
Analysis shows that “millions on vibecoding” is more of a myth than a reality, if we consider the net income from generated code. The success of projects lies in the ability to quickly assemble an MVP, use third-party infrastructure, and scale the business through effective distribution and customer work, and not just through cheap code. The unit economics of these projects are based on a comprehensive approach. Before believing in another success story, always demand statistics screenshots and ask the main question: how much money will each ruble spent bring. For those who are ready to experiment, we recommend trying different models on the same task. Use promo code CTRLAI for a 20% discount on any plan at SYNTX.AI.
