Credits, tokens, messages: why AI app builder pricing is hard to predict

Lovable credits, Bolt tokens, Replit checkpoints, Base44 message credits, v0 token rates: how each unit is spent and why the cost of an app is hard to estimate.

Tomas·5 min read
0billing units across the builders
0.0–1.7Lovable credits for a single request
0–25×higher per-token rate on v0’s stronger model

Most AI app builders don't charge per app. They charge for AI work in units that change size with what you ask. The same request can cost more on a large project than on a small one. That's why the plan price tells you little about what a finished app will cost.

I'm the CTO at Leanfinit, and I spend a lot of time reading how other builders bill, since our users ask about it constantly. Below is how each unit works as of November 2026, using only the vendors' own documentation.

Five units, one problem

Each builder names its unit differently, but they all meter the same underlying cost: the model processes your request plus the context it needs to read.

BuilderUnitWhat changes the price per request
LovableCreditsRequest complexity
BoltTokensProject size, since files are synced to the model
ReplitUSD credits, billed per checkpointEffort the Agent spends
Base44Message credits + integration creditsScope of the change; live app usage
v0USD credits, consumed as tokensModel used and context size

Credits that scale with complexity

Lovable charges a flat 1 credit per message in Plan mode. In Build mode the cost depends on the request, and its examples run from 0.5 credits ("make the button gray") to 1.7 for a landing page with images. To its credit, Lovable shows what each message cost in the message history (source). Cloud, AI gateway and connectors are billed separately by usage.

Base44 uses two separate pools. Message credits pay for building, and the docs say plainly there is "no set credit amount used per message": roughly 0.5 for a simple change, 2 for a complex module, 3–4 for app-wide changes. Integration credits pay for what the live app does: sending email, generating images, calling an LLM, running backend functions (source). This second pool is the one people forget. If it runs out, the app's integrations fail and your users see a generic error.

Tokens that grow with your project

Bolt bills tokens, input plus output, and most of them go on syncing project files to the AI. Its help center puts it directly: "the larger the project, the more tokens used per message" (source). The same one-line request costs more in month three than in week one, because the model reads more code each time.

v0 converts USD credits into tokens at a rate that depends on the model. As of November 2026, v0 Mini is $0.20 per 1M input tokens and $1.20 per 1M output, while v0 Max is $5 in and $25 out (source). Input includes the prompt, uploads, chat history, source files and Vercel context. Picking the stronger model raises the per-token rate about 20 to 25 times.

Effort-based checkpoints

Replit bills paid Agent work by "effort-based pricing based on request complexity", bundling work into checkpoints (source). A complex feature becomes a single checkpoint that can cost more than several small ones, and Plan Mode reasoning can be billed even when no code changes. Replit does ask for confirmation before a paid action starts, and lets you set alerts and hard budget limits (source).

Why the cost of an app can't be quoted upfront

Put these together and you see why no builder can tell you what your app will cost.

You don't know the number of requests. A tracker can take a few prompts or dozens, depending on how often you change your mind and how often the AI gets it wrong.
Each request gets more expensive as the app grows. This is explicit for Bolt and implicit wherever context size drives the price.
Fixes cost the same as features. When the AI breaks something, the fix can be billed like any other request. Base44 is an exception: automatic AI fixes are free on its paid plans.
Running the app can cost money too. Base44 integration credits, Lovable Cloud and Replit's databases and publishing are metered separately from building.

Running out: rollover and top-ups differ

When an allowance runs out, the rules are less consistent than the units. As of November 2026:

Lovable: unused monthly credits roll over while subscribed, expiring two months after issue on monthly plans. Top-ups on Pro and Business are valid for 12 months.
Bolt: paid tokens stay valid for two months. Top-ups ("reloads") exist only on the highest monthly Pro plan or annual Pro plans, and they don't expire.
Replit: Core credits don't roll over; Pro credits roll over for two months. Credit packs for Core and Pro start at $100 and expire after six months. Usage beyond your credits is invoiced unless you set a hard limit.
Base44: unused credits expire at every reset, and no on-demand top-up is described.
v0: monthly credits roll over and expire after 65 days. Purchased credits last a year but need an active paid plan.

How we think about it at Leanfinit

During the beta, Leanfinit is fully free, so none of this arithmetic applies to building with us yet. I won't describe pricing we haven't announced.

What I can say is what we learned from reading these pages. People new to AI tools don't think in tokens, and a unit whose size depends on your project is hard to trust. Some of our beta users click an example chip or send "fitness tracker" as a whole prompt; they have no way to guess how many credits that would take. If we charge later, the unit should be something a non-technical person can understand before they press a button.

For side-by-side numbers, see app builder pricing compared, Lovable vs Bolt credits and low-code pricing models. For what each free plan gets you, read our walk through the free tiers.

You can build for free in the studio: studio.leanfinit.com.