TextSetu vs per-word pricing
Per-word pricing made sense when a person typed every word. Here is what it costs once a machine writes the draft, and what to compare instead.
This page compares TextSetu with a pricing model, not with a company. Per-word invoicing is how most of this industry bills, and it is worth understanding on its own terms before you compare any two vendors.
Where per-word pricing came from
Per-word pricing is a reasonable way to buy human translation. A translator's cost really is roughly proportional to the number of words they read, think about and type. Paying per word means paying for effort, and both sides can check the number.
That logic holds right up until a machine writes the first draft. Then the cost of producing a word collapses, the cost of deciding whether the word is right does not, and the two stop being related at all. The invoice still counts words. The work no longer does.
What per-word billing does to your budget
Three things follow from a per-word price, and none of them are anyone's fault. They are simply what the model implies.
Your bill multiplies by language count. Ten languages is ten times one language, forever, even though the ninth language reuses most of what the first one taught the system.
Your bill moves with decisions other people make. A developer ships a feature with 400 new lines of copy. Marketing rewrites the homepage. Neither of them was thinking about the translation budget, and neither could have told you what their change would cost.
You cannot forecast it. Ask what next quarter costs and the honest answer is a function of release velocity, market count and how much copy gets rewritten: three variables the person holding the budget does not control.
There is a quieter consequence too. When the AI cost is buried inside a per-word price, you never learn what the machine actually cost. That number is knowable, published by the model providers, and small. It is only invisible because the pricing model hides it.
What we do instead
TextSetu charges nothing for the AI. You connect your own Claude, OpenAI, Gemini or DeepSeek account, and the model charges land on your provider's invoice at their published rate. We never mark them up, because we never touch them.
That changes what you are buying. You are not buying words any more; you are buying the workflow around them: the glossary that stops the model renaming your product, the queue where a person approves every draft, the memory that stops you paying twice for the same sentence, and the record of who approved what.
Compare these instead of price per word
If you are evaluating two platforms, per-word price is the least informative number on the page. These questions separate them faster:
- Who pays the model provider? If it is the vendor, the AI price is inside your price and you will never see it.
- What happens to the tenth language? If it costs the same as the first, nothing is being reused.
- What does a one-line change cost? If it re-translates a file, you are paying for work nobody needed.
- Are reviewers charged per seat? If they are, you will ration the colleagues best placed to catch errors.
- What does export cost, and what is the notice period? The easiest platform to start with is the one you can leave.
- Can you see spend while it accrues, or only on the invoice? After the fact is the worst time to learn what a release cost.
Check the arithmetic yourself
The calculator on our pricing page takes your word count, your language count and a model, and shows the whole calculation: words to tokens, tokens to dollars, per language and in total, next to the same volume priced at an industry-typical per-word rate. The rates come from each provider's published pricing page and carry the date they were last checked.
It is an estimate, not a quote, and it is deliberately falsifiable: every step is on screen, so you can disagree with any of them.