How much does a SaaS MVP cost in 2026, with a senior developer augmented by AI?

The short answer: in France, in 2026, a SaaS MVP built by a senior freelancer costs between €8,000 and €22,000 according to Dimitri Dumont, and a simple MVP delivered by an agency between €40,000 and €60,000, for two to three months of development, according to Lonestone. A one-to-five spread, for products that go by the same name.

That spread does not come from the number of screens. It comes from what the product has to guarantee, and from who builds it. That is what you need to understand before comparing two quotes.

What drives the price: scope, not the number of screens

Two ten-screen MVPs can differ threefold in cost. The difference lies in four areas that never show up on a mockup.

Accounts and roles

A single kind of user who logs in and manages their own data is the simplest case. As soon as there is an admin, team members, guests or end customers, every screen and every query has to know who is allowed to see what. That is where most security holes hide, and so a good share of the development time.

Payments

Charging a subscription through a Stripe-hosted checkout page is quick to wire up. Paying money out to third parties is a different job: connected accounts, commissions, refunds, verification duties. On Shooting Pilot, a platform where professional photographers deliver their galleries and get paid by their clients, payments go through Stripe Connect. It is not an option you bolt on at the end: it shapes the data model from day one.

Multi-tenancy

A B2B SaaS sells to companies, and one company’s data must never appear in another’s account. That isolation is decided in the architecture, not in the interface. Done right from the start, it costs a few days; retrofitted later, it costs a rewrite.

AI inside the product

Calling a language model takes an afternoon. What costs is everything around it: making the output explainable, handling model errors, keeping the cost per call under control, deciding where the data runs. In Novera Hire, a recruitment ATS, candidate scoring is transparent: the recruiter sees why a profile stands out. In Novera Recorder, transcription and minutes run entirely on the local machine, because meeting content must not leave it. Both choices weighed more than the model call itself.

What AI changes in the budget, and what it does not

Both sources quoted above agree. According to Lonestone, generative AI cuts development time by 10 to 20%, without a proportional drop in cost. According to Dimitri Dumont, AI has increased the value delivered per day; it has not collapsed the cost of a serious SaaS.

That matches what I see when building with Claude Code. AI clearly speeds up:

• repetitive code: forms, list screens, API routes, migrations;
• tests, which no longer have an excuse not to be written;
• refactoring and documentation, which finally keep up with the code.

It does not replace:

• framing: deciding what the MVP must prove, and what it will not do;
• architecture: permissions, data isolation, payments, where a mistake costs a rewrite;
• security and review: code produced quickly still needs a reviewer who knows what to look for.

In other words, AI makes a senior developer more productive; it does not make a junior developer senior. The budget mainly drops where the work was mechanical.

Three budgets, three different things

Comparing these ranges as if they bought the same product is the most common mistake. They do not buy the same thing:

1. A proof of concept, €5,000 to €15,000 over two to four weeks (Lonestone): it validates a single critical feature. It is not a product you sell.
2. An MVP with a senior freelancer, €8,000 to €22,000 (Dimitri Dumont), i.e. 15 to 40 days of development: one person who frames, codes and ships to production.
3. A simple MVP from an agency, €40,000 to €60,000 over two to three months (Lonestone): a team, with project manager, design and QA. You also pay for coordination.

The right question is not “which one is cheapest?” but “what do I need to learn what I need to learn?”. Sometimes a proof of concept is enough. An agency makes sense when the product needs several disciplines working in parallel.

The costs quotes leave out

• Maintenance: 10 to 25% of the initial cost per year for hosting, updates and changes (Lonestone).
• A security audit, €5,000 to €10,000 when it is required (Lonestone).
• AI model calls, billed by usage: they grow with your users and should be estimated at framing.
• Third-party services (email, storage, payments), which take a cut of each transaction or a monthly fee.

Cutting the bill without cutting the product

• One complete user journey rather than five half-built ones: the one that proves someone will pay.
• A single user role at launch, if the business allows it.
• A hosted checkout page rather than a custom flow, until conversion is measured.
• Proven building blocks for authentication and storage: that is not where your product stands out.
• Usage tracking from launch day, so the rest of the budget goes to what people actually use.

This is the method I apply to my own products: nine so far, listed on the projects page. Framing comes first, because framing is what sets the price. The full method: how I build a SaaS solo with Claude Code.