What vibe coding costs after launch: credits, fixes and the bill at 100, 1,000 and 10,000 users

Every post about what vibe coding costs is about building: credits per message, tokens per day, the plan that turns the agent on. This one is about the month after launch, when the app has customers and the bill has lines on it you did not pick. Here is the arithmetic for a Lovable app with Supabase, Resend and Stripe at 100, 1,000 and 10,000 users, with every assumption stated so you can argue with it, and every price read from the vendor's own page on 27 October 2026.

The short version: the subscriptions are small and nearly flat, from $50 a month to $119 depending on who holds teams and plans. The bill that moves is elsewhere, in three places. Card fees, which scale with revenue and pass every subscription by 1,000 users. Organizations, which one vendor prices per team and which turn $0 into $900 a month at 1,000 customers. And fixes, the credits and hours spent keeping generated code alive, which no pricing page shows and which is the largest line at every size below 1,000 users.

The assumptions

Change any of these and the total moves. That is why they are listed.

• A B2B app built on Lovable Pro ($25 a month for 100 credits, read on 27 October 2026), published on Lovable's hosting with a custom domain, which the Pro plan includes.
• Data in Supabase, on the Pro plan ($25 a month, 100,000 monthly active users included, then $0.00325 each) from the day a stranger signs up. The Free plan pauses a project after a week of inactivity and is for trying things, not for customers.
• Transactional email through Resend: free for 3,000 emails a month and 100 a day, Pro at $20 for 50,000.
• Payments through Stripe at its published US card rate, 2.9% plus 30 cents, with the 0.7% Stripe Billing fee shown separately for the case where Stripe runs the subscriptions. Lovable's built-in payments offer Paddle instead, at 5.0% plus 50 cents (a flat 10% under $10), as a merchant of record that handles sales tax.
• Customers are teams of ten: 100 users is 10 customers, 1,000 users is 100, 10,000 users is 1,000. Each customer pays $49 a month on one card. Revenue is therefore $490, $4,900 and $49,000 a month.
• The account layer, meaning sign-in, teams, roles, plans and limits, is priced three ways: Lovable's built-in sign-in with the teams and plans the agent writes; Clerk for sign-in and organizations; or BuildBase.
• US cards only. International cards add 1.5% on Stripe.

The subscriptions

The builder is $25 at every size, and that is the first honest surprise: the plan you built on is also the plan you host on, and a Lovable app serving 10,000 users costs the same subscription as one serving ten. What changes is the credits, which is the fixes section below, and Lovable's own page shows Business at $50 if you need more of them.

The database is flat too. Supabase Pro's 100,000 included monthly active users cover all three sizes, and the $0.00325 per user beyond that would not start until the tenth customer of the tenth thousand. Storage and bandwidth above the plan's allowances are extra and are not in this table, because at these sizes a typical B2B app does not reach them; yours might.

Email is the line that steps. Ten customers send a few hundred emails a month and sit inside Resend's free tier; a hundred customers pass the 100-a-day cap on the first busy morning, so the $20 plan goes in at 1,000 users and stays.

The account layer is where the three stacks part ways, and it is the whole argument, so it gets its own section after the fees.

At 1,000 users the BuildBase column adds to $89 a month: $25, $25, $20 and $19. That is the number on the card, and it is the number to hold in mind while reading the fees table, because the fees pass it.

One line that is deliberately absent: hosting. Lovable's plan includes it, which is the quiet advantage of building where you publish. The same app on v0 is a Next.js app on Vercel, where the Hobby plan is for personal, non-commercial use, so a product needs Pro at $20 a seat; on Bolt, hosting is inside the $25 with a request and bandwidth allowance per account; on Replit, publishing is billed by usage from the plan's credits. None of those is large at these sizes, and all of them are a line the Lovable table does not have.

Another that is absent on purpose: the model bill, if your app has an AI feature of its own. It is the one cost that scales with usage rather than with users, it varies by a hundred times between apps, and the right way to carry it is to charge for it, which is what credits and usage limits is about. Any number here would be invented.

The fees, which scale with revenue

The arithmetic for the first row: 2.9% of $490 is $14.21, plus ten charges at 30 cents is $3.00, so $17.21. At a hundred charges, $142.10 plus $30; at a thousand, $1,421 plus $300. Stripe Billing is 0.7% of the same revenue when Stripe runs the subscription itself rather than your code charging cards. Paddle's row is 5.0% plus 50 cents a charge: $24.50 plus $5 at ten customers, and so on.

Two readings. First, by 1,000 users the card fees are larger than every subscription in the first table combined, which is the sense in which the subscriptions are not the expensive part. A founder who has argued for an hour about a $19 plan and then accepted 2.9% without reading it has the proportions backwards by a factor of nine at 1,000 users, and thirty-five at 10,000 against Grow's $49. Second, the Paddle row is not a mistake to pay: Paddle is the merchant of record, which means it registers, collects and remits sales tax in the places you sell, and at ten customers $12 a month over Stripe is a fair price for never thinking about VAT. At 1,000 customers the same choice is $1,229 a month, and most companies have hired for the tax question by then.

BuildBase's billing is bring-your-own Stripe at Stripe's rates with a 0% platform fee, which is why its row is zero: the fees in the first row are the same whether your app or the platform created the subscription.

The line nobody puts on a pricing page

A vibe-coded app is not finished when it launches. It is finished when it stops needing fixes, and nothing on the three vendors' pricing pages prices that. Two ways to count it, both with their assumptions showing.

In credits. Lovable's pricing page gives examples: 0.50 credits to make a button grey, 1.20 to add authentication, about 1.70 for a landing page with images. Pro's 100 credits are therefore plenty for building and thin for a loop, and a loop is what the reviews describe: across the builders, about 63% of the roughly seventy negative Trustpilot reviews we read in October 2026 were about credits, tokens or charges spent on fixes that did not fix. On Replit, a founder reported $607.70 in charges beyond his plan in the week of July 2025 that his agent deleted a production database, and a one-star review from 20 September 2026 says that "if you set the limit 50 dollar it turns out take 300 dollar". The honest accounting is a range: fixes fit inside the plan in a good month, and in a bad one they cost a second plan or an overage you did not budget.

In hours. Our maintenance arithmetic for AI-authored auth code assumes an agent scaffold costing about $2,400 in engineer time at $150 an hour and maintenance at 40% of that in year one and 65% in year two, which is $960 and then $1,560 a year, or about $80 to $130 a month. That is for the sign-in alone. Teams, roles, plans and the webhook are each a second copy of the same curve, and the audit record for generated versions of them is what the five checks is about.

Put the two together and the fixes line, at 100 users, is somewhere between $0 and a few hundred dollars a month, against a subscription total of $50 to $69. It is the largest line on the bill at every size below 1,000 users, and the only one you cannot read off a page.

There is a way to make the line smaller, and it is not a vendor. It is the number of things in the app that can be wrong. A role column the agent wrote is a thing that can be wrong, and so is the page that checks it, the webhook that sets the plan, the cron job that resets the monthly limit and the dev route that grants credits. Every one of those that lives in a platform instead of your repository is a fix you will not pay for, in credits or in hours, and that is the only sense in which the bundle's $19 is cheap: it is the fixes line, prepaid and flat.

The account layer, which is where the stacks part ways

Built-in sign-in plus agent code: $0. Lovable Cloud's sign-in is free with the plan, its payments take a subscription per user, and everything after login (a workspace per customer, roles, a plan that changes what the app does, seats, limits) is code the agent writes. The $0 is real. The cost is the fixes line above and the risk register in the five checks post, and it compounds with every customer, because every table the agent adds for teams is a table with a rule to get right.

Clerk: $0, then $900. Clerk's user pricing is generous: 50,000 monthly retained users free on every plan, $0.02 each beyond that, and Pro at $25 if you want the branding gone and the add-ons. Organizations are priced per team: 100 free, then $1 a month for each one from 101 to 1,000, with volume rates after. So the line is $0 at 10 and 100 customers and $900 at 1,000, which is the number that surprises people, because the user count never got near a limit. Plans, limits, seats and credits are still yours to build on top, or Clerk Billing at 0.7% of volume for the subscription part.

BuildBase: $19, then $49. Launch is $19 a month with 25,000 monthly active users, 1,000 workspaces and three team members included; Grow is $49 with 100,000 users and 25,000 workspaces. The 1,000-workspace line on Launch is exactly the 10,000-user case, so the table budgets Grow there rather than sitting on a ceiling. There is also a Free plan at $0, with 10,000 users and 50 workspaces, and the 100-user case fits inside it; the table budgets Launch anyway, because Free keeps the BuildBase badge and our sign-in domain, which a launched product usually wants gone. Yearly billing is ten months for twelve, and the paid plans start with a seven-day trial without a card, read from our own pricing constants and checked against the live page on 31 October 2026. What the line buys is the whole layer the other two leave to the agent: sign-in, workspaces, roles enforced by the API, plans with versions, feature flags, credits, the lifecycle email, and the webhook that moves a plan.

The teams column is the one to stare at. At 1,000 customers the three stacks read $0, $900 and $49 a month, and the $0 is the one with the fixes line attached.

A worked example of what "per team" does, since the shape surprises people more than the number. Suppose the app grows from 100 customers to 1,000 over a year. The built-in stack's invoice does not change. Clerk's organizations line goes from $0 to $900 a month in steps of $1 per new customer from the 101st, so the first customer in month two costs $1 and the last one in month twelve costs $1, which is simple to reason about and adds up to $10,800 a year at the end. BuildBase's line goes from $19 to $49 in one step, when the workspace count crosses 1,000. Neither is wrong. Clerk is pricing the organization as the unit, which is honest about what a B2B app is made of; the bundle is pricing the plan. The question for your spreadsheet is whether your customers are teams, and if they are, which of the two shapes you would rather explain to a finance person in month twelve.

Where the built-in stack wins

Under about a hundred customers, selling to individuals rather than teams, with no plan that has to change what the app does, the built-in stack is the right call and the cheapest by $19 a month. Lovable's sign-in ships SAML SSO and OIDC, which BuildBase does not, so an app whose buyer signs in with Okta is better served there. Paddle's row is a tax department for $12 a month at that size. And one fewer vendor is one fewer thing to explain to an agent.

Clerk wins the pure sign-in case at scale: 50,000 free users is more than most apps ever have, and if your customers are individuals, not teams, the organizations line never appears.

Where the bundle wins

The moment customers are teams and plans have to do something. The organizations line is the visible version: $49 against $900 at 1,000 customers. The invisible version is the fixes line, which the bundle turns from a curve into a flat subscription, because the roles, the plan checks and the webhook are not code in your repository. And 0% on billing means the fees table is the same table whichever way the subscription was created.

What moves the total

• Price per customer. Every fee row is a percentage of revenue. At $29 a customer the card fees fall by about 40%; at $99 they double. The subscriptions do not move.
• International cards. Stripe adds 1.5% for cards issued outside the US, so a European customer base turns 2.9% into 4.4% before currency conversion.
• Supabase beyond the plan. More than 100,000 active users, or more storage and bandwidth than Pro includes, adds usage lines; the per-user rate is $0.00325.
• The builder. Bolt Pro is $25 with 10M tokens; Replit Core is $20 with $20 of credits and usage-billed publishing on top; a v0 app on Vercel needs Pro at $20 a seat, because the Hobby plan is for non-commercial use. Swap the builder line accordingly.
• More fixes. Lovable Business at $50 if 100 credits a month is not enough, or the overage on Replit. This is the line that has no ceiling.
• Seats on your side. Launch includes three team members in the BuildBase console; Clerk Business adds dashboard seats at $20 each. If your own team is larger than that, the next plan up applies.

Annual billing, and the discounts that are real

Every vendor here sells a yearly price, and the discounts are not the same size. Lovable Pro falls from $25 to $21 a month on annual billing, read from its subscription plans page today. Clerk Pro goes from $25 to $20, Replit Core from $20 to $18, and BuildBase's yearly price is ten months for twelve, so Launch is $190 a year and Grow $490. Supabase, Resend and Stripe do not discount for a year's commitment. On the 1,000-user BuildBase column that turns $89 a month into about $82 if you pay the builder and the account layer annually, and the fees do not move at all, which is one more way of saying where the money is.

How to redo this for your own numbers

The method is four lines in a spreadsheet, and it is worth keeping the spreadsheet rather than the answer. Put customers, users per customer and price per customer at the top; everything below is a formula. Subscriptions are a lookup against each vendor's tiers at your user and organization counts. Fees are a percentage of revenue plus a per-charge amount times the number of charges. The account layer is three columns, and the one that reads $0 gets the fixes estimate from the section above, in hours at your rate. Then change the price per customer and watch which lines move. The ones that do not are the ones you have been arguing about; the ones that do are the ones to argue about instead.

Re-check these numbers

Prices here were read on 27 October 2026 from each vendor's own page: Lovable's subscription plans page, Supabase's pricing page, Resend's pricing page, Stripe's pricing page, Lovable's payments documentation for the Paddle rate, Clerk's pricing page, Vercel's pricing page, Bolt's pricing page, Replit's pricing page, and our own plan constants checked against the live pricing page. Vendors change prices without notice, and Lovable in particular has changed its plan structure more than once this year. Before this arithmetic decides anything, open the pages again. The assumptions are listed so you can change them; the method survives the numbers.