The three-hour prototype is documented, but the revenue is not a typical outcome
A February 2025 Indie Hackers report says Pieter Levels had never made a game before. Using Cursor, he prompted his way to a playable browser flight simulator in about three hours. The early version ran on HTML and JavaScript, and he quickly added shooting, mobile controls, explosions, multiplayer behavior, and paid aircraft.
The report later updated the story with figures Levels shared publicly. It listed about $38,000 per month from advertising blimps and about $360 from twelve F-16 purchases, for a reported total of $38,360 per month less than three weeks after the project began. That is a founder-reported revenue run rate relayed by an edited community publication. It is not independently audited profit, and it is not evidence that another builder will produce the same result.
Distribution and direct support are part of the causal story
The project received an Elon Musk repost and, according to the same report, reached about 89,000 players with a peak of roughly 26,000 online at once. Levels also entered the experiment with an audience numbering in the hundreds of thousands. A new creator without that distribution cannot treat the revenue figure as the normal consequence of three hours of coding.
The technical support was unusual as well. The report says a Cursor cofounder contacted Levels and helped when he ran into problems. Grok 3 supported backend work, while Claude 3.7 Sonnet and ChatGPT appeared in the public tool list for the project. Builders can copy the order of a fast test, but they cannot assume celebrity distribution, an existing audience, or direct vendor intervention.
Build one experience that a stranger can use today
Define one input, one visible action, and one result that a stranger can understand without a long explanation. Reduce the scope until it can be demonstrated in a day. The prototype does not need a complete world, account system, feature roadmap, or polished brand. Its job is to reveal whether anyone wants to try the core experience.
Then test only one paid item. The flight simulator offered an F-16 upgrade at $29.99 and obtained a small number of real purchases before the larger reported monthly figures appeared. Your price may be lower and your product may not be a game, but the test should remain interpretable. Launching subscriptions, ads, memberships, and multiple tiers at once makes the first evidence impossible to read.
Give code, debugging, and content separate minimal tests
If the prototype uses several models, prepare a fixed sample and acceptance rule for each type of work. Check current public channel status, create an isolated project key or label, and run one minimal real request for code, debugging, or content under a small preset budget. A model appearing in a marketplace only makes it a candidate; it does not prove that the intended key, endpoint, input, and acceptance rule work together.
Record the effective model, response status, visible usage, failure type, retries, and human editing time. Stop when the same failure repeats, correction time exceeds the value of the result, or the budget limit is reached. Reliability does not mean that every request succeeds. It means failure is observable, spending is bounded, and a manual or alternate route remains available.
Count the cost of one completed result
Listed model price is only one component of cost. Failed calls, uncontrolled retries, switching tools, and human repair all belong in the cost of one usable outcome. A $29.99 paid item cannot support a workflow that requires expensive generation and prolonged manual correction. A popular demo is also not a business until customers pay enough to cover delivery and support.
https://APIToken.Company can serve as one controlled multi-model testing entry point. Within the models currently and lawfully offered on the site, review the marketplace and public channel status, create an isolated key, preserve usage records, and set a stopping budget. Current models, prices, groups, and availability follow the live site pages rather than this case article.
Copy one paid test, not the reported income
This article is not an income promise and does not claim that Pieter Levels, the flight simulator, or any tool in the original report used APIToken. The reported $38,360 per month is revenue, not profit. It must be read alongside the founder's existing audience, the celebrity repost, rapid public distribution, and direct technical help.
The minimum action today is to ship one experience a stranger can try, attach one paid item, and show it to an initial group of users. Only after someone commits time or money should you compare models with isolated keys, minimal real requests, visible usage, and a budget-based stopping condition.
Source and evidence boundary
The case facts come from Indie Hackers' February 27, 2025 edited report, Pieter Levels used AI to build a viral flight simulator in 3 hours with no background in game development. The page documents the three-hour prototype, early purchases, reported player counts, the Elon Musk repost, the tool list, and the later monthly revenue update.
This article treats the report as grade B evidence. It preserves the source's timing and amount while identifying the figures as founder-reported revenue rather than audited profit. It also keeps the large existing audience, celebrity distribution, and vendor assistance visible as non-replicable conditions.