AI builders charge for the AI separately from the subscription, in a unit of their own: credits per message, tokens, or a dollar balance. Your plan includes an allowance of that unit, every message you send uses some of it, and when it runs out you wait for the reset, buy more or upgrade. On Lovable, for example, Pro is $25 a month for 100 credits, and Lovable's own examples put one build message at 0.50 to 2.00 credits.
A message has no fixed price. What you pay depends on how big each change is, how many tries it takes, how many errors need fixing and, on some tools, which AI model does the work. This guide explains the units each builder uses, why two people's bills for similar projects differ, and how to estimate and control your own spend.
Where the numbers come from: each builder's own pages. Lovable figures are as of 8 October 2026; Bolt, v0, Replit, Base44 and Hostinger figures are as of 18 October 2026. Prices change, so check the linked source before you buy. Side-by-side tables are on our comparison page.
Why AI builders charge in credits at all
Every message you send to an AI builder runs a large language model. The companies that run those models charge per token, the small chunks of text a model reads and writes, so a message that rewrites half your app costs more to run than one that changes a button color. The builder has to pass that cost on somehow.
Most sell an allowance, shown as tokens, as credits or in dollars. The subscription price is what you see in advance; how fast you use the allowance decides whether the plan is enough.
Three units, builder by builder
Credits per message: Lovable and Base44
Lovable counts in credits. The Free plan gives 5 credits a day, up to 30 a month. Pro is $25 a month, or $250 a year, for 100 credits a month plus 5 a day, and larger credit tiers cost more. Lovable's examples put a build message at 0.50 to 2.00 credits. Unused monthly plan credits roll over while the subscription stays active, and paid plans can top up. One detail matters for budgeting: hosting and Lovable's built-in backend also draw on credits after a monthly grant of 20 Cloud credits, and when credits run out the published site stays live but the backend can pause. Sources: plans and credits, subscription plans. Our full Veloxa Build vs Lovable page has the rest.
Base44 uses two kinds of credit. Message credits pay for building: 25 a month on Free (up to 5 a day), 100 on Starter and 250 on Builder. Integration credits pay for built-in services your app itself uses, such as sending email or calling AI: 100 on Free, 2,000 on Starter and 10,000 on Builder. Unused credits expire when they reset. Starter is $16 a month billed yearly or $20 monthly; Builder is $40 yearly or $50 monthly. Sources: pricing, credits.
Hostinger AI Builder also counts in credits, and says each request costs credits according to its size. Top-up credits expire three months after purchase. Source: Hostinger pricing help.
Tokens: Bolt
Bolt shows tokens directly. Free gives 300K tokens a day, up to 1M a month. Pro is $25 a month for 10M tokens with no daily limit, and larger token tiers cost more. On paid plans, unused monthly tokens roll over and stay valid for two months; Free plan tokens do not. Token reloads, which do not expire, are available on the largest monthly Pro plan or any yearly Pro plan. On Teams, at $30 per member a month, each member gets their own token allotment, and tokens are not shared. Sources: pricing, tokens. More on Veloxa Build vs Bolt.
Dollar credits: v0 and Replit
v0 counts in dollar credits. Free gives $5 of credits once and up to 7 messages a day. Plus is $30 per user a month and includes $30 of credits per user, plus $2 of free credits a day on login. Generation pauses when credits run out. Unused monthly credits roll over and expire after 65 days; extra credits bought on paid plans expire after a year. Hosting is separate, on Vercel. Sources: pricing, pricing docs. See Veloxa Build vs v0.
Replit also uses dollars. Core is $20 a month ($18 billed yearly) with $20 of credits, plus a Free Mode allowance for everyday Agent work; Pro includes $100 of credits. Paid work is priced by effort, and Replit asks before a paid action starts. Publishing is billed by usage and draws on your monthly credits first. Pro rolls unused credits over for two months; Replit's pages do not say whether Core does. Credit packs expire six months after purchase. Sources: pricing, AI billing. See Veloxa Build vs Replit.
Dollars at published per-token rates: Veloxa Build
Veloxa Build, our builder, prices AI usage in dollars at per-token rates published on its pricing page. Each plan includes a dollar amount of AI usage a month: $1 on Free (up to $0.25 a day), $6 on Starter ($9 a month), $20 on Pro ($25 a month) and $50 on Team ($59 a month). Paying yearly gives two months free.
- Veloxa Build, used for building and changing your project: $1.75 per million input tokens, $0.325 per million cached input tokens, $5.50 per million output tokens.
- Veloxa Build Fast, used for small tasks such as naming a project: $0.1875, $0.0375 and $0.625 per million.
Those rates are the model provider's list price plus 25%. The builder uses an open-weights model and you can't choose a different one. Every message shows its cost in the builder. The plans also carry a rough guide, such as about 100 to 200 AI edits a month on Pro, but that is guidance, not the unit. Unused AI usage does not roll over, and hosting never uses it: when it runs out, the AI pauses until the next month or a bigger plan, and published sites keep working.
Why two people's bills differ
The size of each change
A token-priced message pays for what the model reads (your request and the parts of the project it needs) and what it writes. As a project grows, there is more to read. That is why Lovable gives a range per message, Hostinger charges by request size and Replit prices by effort.
The number of turns
This is the lever you control most. A vague first message produces a draft that needs many corrections, and each correction is another message. A clear brief with your real content often gets close in far fewer turns.
Fixing errors
When generated code breaks, fixing it is AI work too. On Veloxa Build, the builder checks the preview after every change and fixes errors it finds, and those fixes use AI usage at the same rates as any other message; they are not free today.
Model choice
v0 and Replit let you choose the model, and Base44 does from its Builder plan; Lovable and Veloxa Build don't, and Bolt offers a choice of agent. Where you can choose, the model is part of the price of a message, so check the rate before you switch.
What else draws on the same pool
On some builders the allowance pays for more than building. Lovable's hosting and backend draw on credits after the monthly Cloud grant, Replit's publishing draws on monthly credits first, and Base44's integration credits pay for the email and AI calls your app makes. On Veloxa Build, hosting never uses AI usage.
Seats, daily limits and expiry
Bolt Teams and v0 give each user their own allowance, which teammates can't share. Free plans often have a daily cap: Lovable 5 credits, Bolt 300K tokens, Base44 5 message credits, Veloxa Build $0.25. And rolled-over credit still expires, after two months on Bolt or 65 days on v0.
See the cost of every message: Veloxa Build shows what each message cost, in dollars, as you build. Start building on the Free plan.
How to estimate what you will spend
- Find the unit and the cost of a typical message. Start from the builder's guidance, such as Lovable's 0.50 to 2.00 credits per build message, then build something small on a free plan and note what your first ten messages cost.
- Count the turns. List the pages or features, allow a few rounds of changes for each, and add a margin for error fixes.
- Multiply, then compare with the allowance. For example, Lovable Pro's 100 monthly credits at 0.50 to 2.00 credits a message works out to roughly 50 to 200 build messages, before the 5 daily credits. On Veloxa Build Pro, $20 of AI usage is about 100 to 200 edits by the plan's own guide.
- Add what else draws on the pool: hosting, backend use and your app's own AI or email calls, where they share the allowance.
- Check after the first week. Replace the guess with your real per-message cost.
How to keep spend under control
- Write a complete first brief. Who it is for, what it must do, and your real copy, prices and photos. Our guide on how to write a build prompt shows what to include.
- Group related changes into one message; keep unrelated ones apart so a problem is easy to trace.
- Undo with version history, not with a message. Asking the AI to put things back is a message you pay for; restoring an earlier version is a click.
- Break loops early. If a fix fails twice, restore the last good version and describe the problem differently.
- Watch the meter. Veloxa Build shows each message's cost; Replit asks before a paid action starts.
- Know what happens at zero. v0 pauses generation, Lovable's backend can pause, and on Veloxa Build the AI pauses while sites stay online.
- Pick a plan after a month of real use, not before. If your usage is lumpy, rollover rules matter more than the headline allowance.
Which pricing model suits you
- Busy one month, quiet the next: rollover matters. Lovable, Bolt's paid plans, v0 and Replit Pro roll unused allowance over, each with its own limits. Base44 and Veloxa Build don't.
- You want to top up mid-month: Lovable's paid plans, v0's extra credits and Replit's credit packs allow it. On Veloxa Build, more AI usage means a bigger plan.
- You want to choose the model: v0 or Replit.
- You want the price in dollars: v0 and Replit count in dollar credits; Veloxa Build publishes its per-token rates and shows each message's cost.
- You build as a team: Veloxa Build paid plans have no seat fees and share one allowance; Lovable allows unlimited members. Bolt Teams and v0 charge per member, and Replit Core allows up to 5 builders.
The comparison table puts the plans side by side, and the Veloxa Build pricing page lists every plan, the AI rates and what happens when usage runs out. If you are new to building this way, vibe coding explained covers what to expect.
Frequently asked questions
How many credits does a Lovable message use?
Lovable's own examples put a build message at 0.50 to 2.00 credits, depending on the change. Lovable Pro includes 100 credits a month plus 5 a day. These are Lovable's figures as of 8 October 2026.
Do unused AI builder credits roll over?
It depends on the builder. On the dates we checked, Lovable rolled unused monthly plan credits over while the subscription stayed active, Bolt kept unused paid-plan tokens for two months, v0 kept unused monthly credits for 65 days, and Base44 and Veloxa Build did not carry unused allowance over.
What happens when I run out of credits?
Building stops until the allowance resets, you buy more, or you move to a bigger plan. What happens to a published site varies: on Lovable the site stays live but the built-in backend can pause; on Veloxa Build, published sites, their data and their sign-ins keep working, because hosting never uses AI usage.
Do automatic error fixes use credits?
On Veloxa Build, yes: when the builder spots an error and fixes it, the fix is a message like any other, it uses AI usage at the same rates, and you see what it cost. For other builders, check how their own pricing pages count fixes.
Try it and watch the meter
The quickest way to learn what your projects cost is to build one and look at the numbers. Start building on Veloxa Build with $1 of free AI usage a month, or read the plans and rates first.
Spotted an outdated figure? Email us at support@veloxahost.com and we will re-check it against the builder's own pages.