Why published MVP cost ranges disagree

Published MVP cost ranges disagree because they are counting different products and calling all of them an MVP. In a survey of 300+ web development firms across 31 countries, run in April and May 2026, the most common quote for a basic website and an MVP together is $1,000–$3,000, named by 59.9% of respondents, and 71% put that build at two to four weeks (GoodFirms, Website Development Cost 2026, updated 7 September 2026). The same survey's most common quote for a custom web application is $20,000–$50,000. A separate guide, updated 21 September 2026 from client reviews, says custom software projects on its platform typically cost $10,000–$49,999, and on the same page that the average reviewed project costs $132,480.29 over about 13 months (Clutch, Software Development Company Pricing Guide, 21 September 2026). Someone who searched how much an MVP costs can cite all four figures in good faith. They are not four measurements of one thing.

The question worth answering is which row describes the thing you are about to buy. A midpoint of the four does not. Match the product to the row that was actually asked about, then buy a named deliverable against that product. If you cannot describe the product in one sentence a stranger could accept or reject, you are not in any of the rows yet, and a budget taken from them will be a budget for a different build.

GoodFirms asked agencies, firms, and independent professionals what they charge for typical scenarios. The methodology says the figures are agency-reported ranges, not a sample of invoices a client paid, that respondents run from freelancers to firms of 200+ people, and that some questions allowed more than one answer so the percentages need not sum to 100. The row that produces the number founders repeat is labelled "Basic Website and MVPs." A basic website and a minimum viable product are collected as one answer. The quick-answer table prints "Basic Website / MVP" at $1,000–$3,000 and two to four weeks. The detailed table prints 59.9% of companies in that cost band and 71% on that timeline. The prose rounds the share to 60%. Use the table. A founder who treats $1,000–$3,000 as the market price of a startup product has adopted a website quote.

The rest of the page shows why that quote feels like the market. Of fixed-price work, 33% of agencies quote $1,000–$5,000 and another 30% quote $5,000–$15,000. The key takeaway combines those into 63% of agencies quoting between $1,000 and $15,000, and describes budgets above $35,000 as rare, under 7% combined. That is a real picture of what web firms say they charge for websites. It is a poor picture of what it costs to put a second company on a shared database, which is the build a founder usually means and a website survey usually does not ask about.

Two other rows in the same survey sit closer to a product, and they are still rows about websites. For a custom web application, 35.5% of respondents — the prose rounds this to 35% — named $20,000–$50,000 as the most common range, and the same 35.5% named eight to sixteen weeks. Another 19% named $50,000–$75,000. For what the survey calls a SaaS-based website, 61% named $25,000–$80,000, and 76% named sixteen to forty weeks. Fintech and neobank sites land in that same $25,000–$80,000 band for 54% of respondents, on a sixteen-to-forty-week timeline for 62%. Read the nouns. The instrument is still a website questionnaire: a SaaS-based website, a fintech website. It does not claim to have priced the application those sites describe. The distance from $1,000–$3,000 to $25,000–$80,000 is what happens when one publisher changes the noun. The distance from a marketing site to the product behind it is a further step the survey does not take.

Clutch is a different object, and the page disagrees with itself in print. Hannah Hicklen's guide, updated 21 September 2026, says the figures come from first-party reviews by verified clients. It does not say how many reviews sit behind either summary. What it does print, a few sections apart, cannot be one cut of one series. Custom software projects "typically cost" $10,000–$49,999. The opening rounds that band to $10,000 to $49,000. Lower down, the average cost of a software development project is $132,480.29, the usual timeline is about 13 months, and the average monthly cost is $10,209.39. Thirteen times $10,209.39 is $132,722, about $240 from the stated average, which is what "about 13 months" should produce if the average and the monthly figure are one series. The typical band is a different cut, and the page does not describe the filter. A founder who tells an investor that Clutch says ten to fifty thousand, and a founder who says Clutch says about $132,000, have both read the page. Neither figure is labelled an MVP. Thirteen months is a long engagement to still call minimal. Using either number as the MVP budget imports a custom-software review mix and presents it as the price of a first version.

Country moves the band before features do. GoodFirms' hourly table puts the United States, the United Kingdom, and Canada at $50–$100 an hour, and India at $10–$15. The same $10–$15 bracket is where that table places Pakistan, Bangladesh, Vietnam, and several other South and Southeast Asian markets. The page's own key takeaway draws the split in those words: South and Southeast Asia at $10–$15, the US, UK, and Canada at $50–$100. Clutch's location table, drawn from listed firms rather than from this survey, puts the United States at $50–$99 an hour and India at $25–$49. The same Clutch guide also says that most companies charge $25–$49 and that the average cost to hire a firm is $25–$49 an hour. The two publishers do not agree about India. They roughly agree that a US firm is quoted in the fifty-to-one-hundred band. A range with no country attached is at least two markets stapled together. A bracket in a directory is what that directory's respondents reported for a category. It is not a named firm's price, and it is not ours.

The wage arithmetic that turns a public salary into a labour floor lives in what an MVP build actually costs, and this article does not repeat it. The only use of it here is the boundary the website row cannot cross. A build quoted at $1,000–$3,000 over two to four weeks is not a senior US team for those weeks. It is a small site, a template, an AI builder, or a labour market the quote has not named. If the proposal names four people and a month, the labour article is the check. If the proposal names a website, you are still on this page, in the first row.

AI tooling compressed the bottom of the website market and left the higher rows standing. Of the same 300+ firms, 98% reported using AI somewhere in the work. On projects they classed as AI-builder work, 33.5% estimated $500–$1,500 and 23.7% estimated below $500. The prose rounds that second share to 23%. Cite the table. GoodFirms' summary of the written answers is that AI split the market rather than flattening it: basic websites became cheaper and faster, while custom development, deep integrations, and AI features inside a product held their price or rose. That sentence is the publisher's reading of comments, not a second census, and it matches the tables on the same page. The $1,000 row gained company. The $25,000–$80,000 row did not collapse into it. A landing page produced for under $500 is evidence about that row. It is not evidence about the custom-application row.

Fixed price is the model these firms report most often, and it means something only after the row is chosen. 66.4% report using a fixed price, 56.6% time and materials, 55.3% a hybrid, and 40.8% a monthly retainer. The shares sum well past 100%, which the methodology allows: multi-select questions, percentages that need not partition the sample. Fixed price is the most reported model. It is not evidence that the scope was knowable when the number was written. A fixed number on an unnamed product is how the $1,000 website and the $50,000 application get sold with the same sentence.

The way through that is dull, and it is the only method the tables leave standing. Name the deliverable. Then name the number. Agree the number before the work starts, against that deliverable, and refuse a rate that pretends every shape of work costs the same. That is the whole of our AI development pricing page: no rate card, because a two-week assessment and a twelve-week rescue do not share a formula. The two-week AI readiness assessment is one fixed fee, agreed before the two weeks start, and the roadmap is yours whether or not a build follows. Fixed-scope integration sprints are a named capability — a copilot, retrieval over data you already have, a workflow — inside a product that already exists, in three to six weeks. A rescue runs eight to twelve weeks and is priced after an audit of the system in hand, not from a template. None of those is a menu price for an MVP. Each is a way to stop a directory range from becoming the contract.

Place the build before you ask anyone for a number. Three descriptions cover the rows above.

If what you need is a site that explains a product, you are in the $1,000–$3,000 row, the timeline is a small number of weeks, and an engineering partner is the wrong purchase. The survey's AI-builder band, with a third of respondents at $500–$1,500 and nearly a quarter under $500, is the same decision with a cheaper tool. Take it.

If what you need is accounts, one workflow, and a stranger who can sign up, you are closer to the custom-application row: a modal quote of $20,000–$50,000, with a further 19% of respondents at $50,000–$75,000, and a timeline measured in months. The acceptance test has to be one sentence both sides can fail. How to scope an MVP so it ships is that decision, applied before the quote rather than after the argument. A price asked before the sentence exists will come back as one of the Clutch figures, chosen because it fit a spreadsheet.

If a second company will share the system — subscription billing, roles that must not leak, an assistant trained without seeing another tenant's rows — you have left the website survey. A multi-tenant agency operating system had to hold subscription billing, role-based access for agencies, admins, and team members, and an admin-trained knowledge assistant in one workspace, because the permission model and the assistant constrain each other. That is a description of scope. It is not a headcount and not a duration. It is also not a basic website. A quote that prices that shape off the $1,000–$3,000 row is on the wrong table. A quote that prices it off the $132,480 average is borrowing someone else's thirteen months.

The build range still leaves the year after unpaid. GoodFirms' phase table, again the most common answer rather than a mean, puts back-end development at 30–40% of the budget (36% of companies), front-end at 20–30% (53%), QA and testing at 10–20% (48%), and discovery and planning at 5–10% (53%). That mix is the build invoice. On retainers, nearly 30% of firms price $500–$1,500 a month and another 28% price $1,500–$3,000, which the page describes as maintenance, content updates, and minor iteration rather than an active build. Separately, and this part is the publisher's planning guidance rather than a respondent share, the same page says to plan 15–20% of the initial build per year for maintenance, and that a two-year total often runs 150–200% of the first invoice once hosting, upkeep, and added features are included. Use the retainer brackets as what firms said they charge. Use 15–20% and 150–200% as a rule of thumb the publisher printed, not as a measured share of the 300+. Either reading has the same consequence. A budget that stops at the build range has planned the invoice and missed the run rate. Where the product includes a model, token cost and evaluation sit in a different lane on purpose, in what it costs to build an AI agent, so a website survey is not asked to measure them.

There is a case where the right purchase is nothing from a product firm. You need a page, a form, and a place to send people. You can name every screen. There is no second tenant. A wrong answer costs an afternoon. Stay in the cheap row on purpose. Paying a startup software development company to produce a brochure turns a directory range into an agency invoice without turning it into a product. The work that company is for starts when the row is the product: architecture, evaluations, a security posture a buyer can read, and a defined end. If something is already in users' hands and you want the failure modes named before you buy any of that, the free production-readiness diagnostic reads a description of what you shipped and returns the ones most likely to bite first.

On Monday, write three lines. The row you believe you are in, copied from a page you can link, not from memory. The country whose hourly band the quote implies. The one sentence that would count as done. If the third line will not stay a sentence, the range is not yours yet, and the next purchase is a bounded scope — the two-week assessment — rather than a build sold off an average. Every dollar figure in this article comes from GoodFirms' April–May 2026 survey, as published on the page updated 7 September 2026, or from Clutch's pricing guide as updated on 21 September 2026. You can open both and check them. None of them is our rate.