Credits, words or seats: how AI writing tools charge, and what each model costs you
By Thomas Lovas Lokoy · Published 16 October 2026 · 5 min read
Pricing pages for writing software are hard to compare because they do not measure the same thing. One sells words, one sells seats, one sells credits, one passes through the cost of the model underneath. This guide describes each model, the way each one goes wrong for a buyer, and the one calculation that lets you compare them all. It names no competitors and quotes no prices other than AI Orbit's own, read from its pricing page on 16 October 2026.
The four models
Per word. You buy a monthly allowance of generated words, and every draft, including the ones you throw away, counts against it. The model is intuitive and it punishes exactly the behaviour a good writer has: regenerating, trying a second angle, asking for a longer piece. A word allowance also makes length a cost decision rather than an editorial one.
Per seat. You pay per person with a login, usually with "unlimited" generation subject to a fair-use clause you have not read. This suits a team that generates constantly and is poor value for a solo publisher who needs eight articles a month. The failure mode is the fair-use clause: unlimited is a marketing word, and the real ceiling is in the terms.
Credits. You buy a monthly allowance of units, and each action costs a stated number of them. Done well, this is the most transparent model, because the price of an action is a small integer you can see before you take it. Done badly, it is the least transparent, because the credit cost of an action can be changed without changing the headline price, and features can be priced in fractions of a credit that are hard to reason about.
Model pass-through. You bring your own key for the language model and pay its provider by the token, plus a fee for the software. This is the cheapest per article at volume and the most unpredictable, because a token is not a unit anyone thinks in and a long research run can cost many times a short draft.
How AI Orbit's credits work
AI Orbit uses the credit model with deliberately few rules. One article written with Sonnet costs one credit; one written with Opus costs three. Length does not change the price, so a 300-word piece and a 4,000-word piece on the same model cost the same. Keyword research, the licensed cover image search, series planning and the keyword-brief preview carry no separate charge. A new account gets three credits that never expire and one free article a day on top; the paid plans run from $19 a month for 100 credits to $149 a month for 1,500, with VAT added in the EU.
Two details of the implementation matter to a buyer. First, the charge is taken before the model is called, and every failure after that point refunds exactly what was charged, including the day's free article; an interrupted run is closed and refunded after twenty minutes of no progress. Second, the monthly refill is a reset to the plan's figure rather than a top-up, so unused credits do not accumulate across months. Both are set out in the credits tutorial, with the full table of plan ceilings.
The calculation that compares them all
Whatever the model, convert it to the cost of one published article, which is the only number that matters to a publisher:
- Drafts per published article. How many generations do you run, on average, to get one piece you publish? For most people it is between one and two: the first draft plus the occasional regeneration with a sharper brief. Be honest; if you regenerate three times, the per-word model costs triple what the pricing page implies.
- Monthly price divided by the allowance, in whatever unit the tool uses, gives the cost of one generation. For per-word plans, use your real average length. For seats, divide the seat price by the number of articles you actually publish.
- Multiply 1 by 2. That is the generation cost per published article.
- Add your review time at your hourly rate. For an AI draft that time is not optional; it is the step that makes the piece publishable. Twenty minutes at whatever your hour is worth will usually exceed the generation cost by an order of magnitude.
On AI Orbit's Pro plan, step 2 is $39 divided by 250 credits, about 16 cents for a Sonnet article or about 47 cents for an Opus one; with 1.5 drafts per published article, step 3 is roughly 23 cents or 70 cents. Step 4 is where the real money is, and it is the same for every tool on this list.
What to check in any pricing page
- Is the unit stated as an integer you can see before acting? A credit cost shown on the button is checkable; a token estimate is not.
- What happens to unused allowance? Rollover, reset or expiry. None is wrong; not knowing is.
- Are failures charged? A tool that bills for a draft the model never finished is passing its reliability problem to you.
- What does the next tier buy? Usually a ceiling: longer articles, more topics per run, a better model. If you will not use the ceiling, the tier is not for you.
- Can you leave with your work? Export in a plain format, cancellation without a conversation.
Which model suits whom
A solo publisher with a weekly cadence is best served by credits or a modest per-word plan, because the volume is low and predictability matters more than unit cost. An agency producing for several clients needs either a higher credit tier with batch runs or a pass-through arrangement with the discipline to watch the bill. A team where many people draft occasionally is the only case where seats are a natural fit, and even then the fair-use clause is the contract.
The cheapest tool is the one whose per-published-article cost, including your review time, is lowest, and that is rarely decided by the headline price. Work the calculation above on your own numbers; the Free tier lets you measure the drafts-per-article figure before you pay for anything, and how it works explains what a credit buys in two minutes.