Why this costs money.

Short version: every job you run here costs us real money, nobody is covering our losses in exchange for a share of you, and we would rather charge you honestly than find a clever way to make you the product. The numbers are below — all of them.

Unfussed charges £7, and quite a lot of the internet will do something that looks similar for free. So this page exists to answer a fair question: why doesn't this? The long answer has numbers in it, and we think the numbers are the honest part, so here they are.

What a job actually costs us

When you run a job, we buy computing from Anthropic, the company that makes the writing models we use. They charge us per word read and per word written — metered, like electricity. Here is what the jobs on sale today cost us to run, in our money:

The jobA typical run costs usThe most it can cost
Fix a marketplace listing0.6p2.4p
A quote for a customer1.6p5.3p
A eulogy2.2p7.9p
A wedding speech2.4p8.5p
A CV rewrite2.7p12.0p
A retirement speech4.5p16.1p
Comparing two builders' quotes4.7p17.8p
Wedding vows5.3p17.8p
A tender response — the dearest run we sell, at its worst5.2p21.7p

The dearest job in the catalogue, with every field filled to its maximum, costs us 21.7p — about 3% of your £7. We are not embarrassed by that gap, and we would rather you saw it than wondered.

That is only the model bill. Around 30p of every £7 goes to the card company before we see it. When a run fails — and models do fail — we pay the full model fee and you pay nothing, because we never charge for failures. Then hosting, storage, and the occasional human answering email.

So why is everyone else free?

Because someone else is paying, and they expect something back. Most free AI products run on investment money — funds covering today's losses in exchange for a share of whatever the product becomes. That model needs enormous growth, and enormous growth needs you inside the product as often as possible. So “free” usually resolves to one of three prices that never appear on a receipt: your attention (ads), your data (profiles, training material), or a subscription you will be steered into once the free tier has taught you to rely on it.

None of that is sinister, exactly. It is a business model — one we did not take. Nobody is covering our losses while we chase scale, which has one awkward consequence: we have to charge you. And one rather nice one: you are the customer, not the crop. We have no ads to serve, no engagement targets, and no reason to want you here one minute longer than your job takes. £7, do the thing, leave. Come back when something else needs writing.

Where your £7 goes

The model bill
Between 0.6p and 21.7p per run, depending on the job and the writer you choose — the table above, plus the tiers below.
The card company
About 30p of each £7, taken at the till. Unavoidable, and not ours.
Failed runs
Ours entirely. A failed run costs us the full model fee and costs you nothing.
Keeping the lights on
Hosting, secure storage for the 30 days your job exists, and a human when you need one.
What is left
The margin that keeps a small, unfunded business alive — the part that answers the email, fixes the bugs, refunds the duds, and means we are still here next year without ever needing to sell what you told us.

The four writers, honestly

Every job here is written by an AI model, and models genuinely come in strengths — measurably better judgement, at genuinely higher prices to us. We use four, and unlike most products we will tell you exactly which, and exactly what they cost us:

Everyday
Claude Haiku 4.5 · $1/$5

Quick and dependable, for jobs with a clear shape. Our simplest jobs are written here as standard.

Standard
Claude Sonnet 5 · $3/$15

The writer most of our jobs use. Good prose, sensible judgement. Most people rightly stop here.

Considered
Claude Opus 5 · $5/$25

Takes more care. Usually better when the situation is tangled or unusual.

Our best
Claude Fable 5 · $10/$50

The strongest writing model money buys — including ours. For the pieces that really matter.

Those are the published rates per million tokens, input and output. A token is roughly three-quarters of a word.

On jobs that offer the choice, you pick the writer before you pay. Here is the same job — a marketplace-listing fix — at all four strengths, our bill beside your price:

WriterCosts usCosts you (typical)Never more than
Everyday — included0.6p~28p of your £740p
Standard1.9p~84p£1.20
Considered3.2p~£1.40£2
Our best6.3p~£2.80£4

Notice the shape of that table: when our bill goes up by a multiple, your price goes up by the same multiple — not more. A stronger writer costs us three times as much, so it uses three times as much of your £7. Stepping up buys a better writer; it is never a fine, and it is never the “real” version of something we were holding back. Every job's included tier is one we are happy to put our name to — most of the catalogue is written at Standard or Everyday because for most jobs that is simply the right tool.

And what stepping up does not buy: a guarantee. A stronger writer usually handles tangled briefs better — the blended family, the delicate apology, the unusual set of facts. It is not a promise you will prefer its draft. If you are unsure, start at the included tier; you can re-run the same answers on a stronger writer afterwards, you will see the price first, and you keep every version.

What we promise

Unfussed is self-help software. Jobs produce drafts and general information from what you supply — you review, you decide, you send. The per-job figures above are computed from each job's own worst-case and typical token counts at the model prices we actually pay, converted at $1 = £0.7388; re-run npm run audit:ceilings against the live catalogue to reproduce them. Model prices are Anthropic's published rates and may change — if they do, our prices move with them, in the same direction and by the same multiple.

Appaya Ltd, registered in England & Wales, company number 16078943. Registered office: Flat 15 Alina Court, 58 Lomond Grove, London, England, SE5 7LE. Read the terms.

Last updated 19 August 2026.