What a 3D website actually costs

What the market publishes

Unlike most of this trade, 3D web work has figures out in the open. These are all published by studios or by directories that verify them, not estimated by us:

• A genuine 3D or WebGL experience: $20,000 to $200,000 and up.
• A product configurator in three.js: $15,000 to $60,000.
• Adding real 3D on top of an existing site build: $15,000 to $80,000.
• A bespoke site with no 3D, mid tier: $20,000 to $50,000.
• Upkeep for an interactive site: $95 to $195 a month for a care plan, and $500 to $5,000 with an agency retainer.

Directories give the other half of the picture, because Clutch publishes every agency’s minimum and hourly rate whether they like it or not. In London, KOTA lists a $25,000 minimum at $100–149 an hour. In the United States, Baunfire lists $75,000, Clay $50,000 and DD.NYC $10,000, all three at $150–199 an hour.

One number in there is worth stopping on. The hourly floor of the American top tier, $150, sits above the ceiling of the British one, $149. It is not a difference in quality. It is a difference in market, and it means the same brief priced in New York and in London comes back meaningfully apart before anyone has discussed the work.

Timelines are published too, and they cluster: 3 to 8 weeks for a 3D piece on a homepage, 8 to 16 for something more involved, 16 to 28 for a complex build.

Why three studios quote the same brief three times apart

This is the part that makes people think somebody is overcharging. Usually nobody is. The brief was open enough to describe three different projects, and each studio priced a different one.

The classic case is a single line: "we want a 3D product on the homepage."

One studio reads that as a model you already have, dropped into a viewer, spinning. Two weeks. Another reads it as a scene with lighting and materials built to match your photography, that has to look right on a four-year-old Android. Two months. A third reads it as something a visitor can configure, which drags in state, options, pricing and a way to save the result. Four months.

All three answered the brief. The brief was the problem.

The six decisions that actually move the number

In rough order of how much they move it.

• How much of the scene is bespoke. A model you already own, prepared for the web, is a different project from one modelled from nothing. This is the single biggest swing and it is usually the vaguest line in the brief.
• Whether the visitor controls anything. Watching a thing rotate and configuring it are not the same build. The moment there are options, there is state to manage, combinations to test and a result somebody expects to be able to keep.
• The performance floor you insist on. "It should work on anything" is expensive, because the work is in the fallbacks, not the scene. Naming the oldest device you actually care about can cut weeks.
• How many distinct screens need designing, which is not how many pages. Twenty product pages on one layout are one design.
• Whether the art direction exists. Applying a system you already have is faster than inventing one. If your brand has no rules for how images and colour behave, that gets decided during the project and it takes time.
• Everything that is not the 3D at all: a content manager so you can edit it yourself, extra languages, integrations with what you already run, and a shop, which brings payments, shipping, stock and tax with it.

The one that catches people out: the file you already have

Almost everyone with a physical product says the same reassuring thing. "We have the 3D files, so that part is done."

It usually is not, and it is worth understanding why before it turns up as a line on a quote.

A file built for engineering or manufacturing describes the object completely, because that is its job: every thread, every internal component, every tolerance. A file built for the web describes only what somebody will see, at the size they will see it. Handing over the first and asking for the second is handing over source material, not a finished part.

The work is real. Geometry has to be rebuilt at a sane density, materials remade, because engineering materials carry no visual information, and the whole thing checked at the size it will actually appear. On a complicated product that is days, sometimes weeks.

None of which means your file is useless. It means "we have the files" moves the number less than you would hope, and it is better to find that out now than in week three.

What it costs after it launches

The published maintenance figures above are for interactive sites generally. A 3D scene sits at the higher end of them, and the honest reason is not that something is fragile.

It is that a scene depends on things you don’t control and that change without asking. Browsers update. Graphics drivers update. A phone released next year renders slightly differently. None of that breaks a page of text and all of it can affect a scene.

So a 3D site needs looking at a few times a year, where a text site can genuinely be left alone. Budget for that from the start. A scene nobody has opened in two years is how a good project turns into an embarrassment.

Why we don’t publish our own figure

Fair question, given everything above.

Because we would have to publish a number wide enough to be true, and a number that wide tells you nothing. The bands at the top of this page span ten to one. Ours would too, and you would still have to write to us to find out where in it you sit.

What we do instead: you tell us which of the six decisions above weigh in your case, and you get a fixed quote with the scope in writing. Not a range. A number, with what it includes and what it does not.

And if you would rather sanity-check us against the market first, everything in this article is published by somebody else. Use it.

When the answer is that you don’t need this

We build these, so it is worth saying where the money is better spent elsewhere.

If your product is genuinely flat, 3D adds cost and maintenance and buys you decoration. A consultancy, a law firm, most software: there is nothing to rotate.

If your traffic is mostly on old phones and poor connections, the constraint is real and it changes the answer.

And if the site is underperforming because nobody can tell what you sell, no amount of three-dimensional work fixes that. It makes the confusion more expensive to load.