A team writes a 40 word caption, decides it needs to run in five languages, and pulls up a translation vendor’s pricing page for a number before committing. The page quotes a rate per word or per character. That rate was built for briefs, product descriptions, and support articles, not for a caption this short, and the number that lands on the invoice is often set by something else on the same page: a minimum project fee.
The number on the pricing page and the number on the invoice
Translation vendors publish rates per word or per character because that scales predictably across a 2,000 word document. A caption of 30 to 50 words is a different kind of job. It is short enough that the per word rate, multiplied out, produces a number too small for a vendor to process profitably, so many pricing structures set a floor underneath it. That floor, a minimum project fee or minimum order size, decides what a caption costs once the job falls under the threshold where per word math stops mattering. The rate on the page is real. It is just not the number that applies here.
This matters more once the caption needs several languages at once, because whatever sets the floor for one language pair can repeat for every pair separately. Before running that arithmetic, it helps to know exactly what each vendor states, and what it does not.
What DeepL Pro, Gengo, and Transifex publish
Each page was fetched directly on 2026-09-09. What follows is only what could be read as text on the page. Prices verified 2026-09-09, not Watchdog data and not on the network’s re-verification schedule.
| Vendor | Billing basis stated on the page | Minimum project fee or minimum order |
|---|---|---|
| DeepL Pro | The page lists subscription plan names for the consumer product (Individual, Team, Business, Enterprise) and the API (Developer, Growth, Enterprise API). It loads as a client rendered page and does not expose a per character or per word dollar rate, or a character allowance, as readable text. [EVIDENCE NEEDED: the per character rate and allowance for each DeepL Pro tier] | [EVIDENCE NEEDED: whether any DeepL Pro tier carries a minimum project fee, since the rate itself could not be confirmed] |
| Gengo | Per word, billed by language pair, with two rate columns per pair, Standard and Business, and separate API and web rates within each. Confirmed rates range from $0.03 per word (API, Standard, easier pairs such as English into British English) up to $0.16 per word (web, Business, harder pairs such as English into Danish or Norwegian). | No minimum order or minimum project fee is stated anywhere on the page. |
| Transifex | Subscription tiers: Starter from $135 a month billed annually ($160 monthly), Growth from $200 a month billed annually ($236 monthly, includes 60,000 AI words a year on the annual plan), Enterprise+ custom priced. Volume beyond the included allotment is sold as “AI words” and “MT words” add on packages, tiered so buying more lowers the cost per word. The page states this tiering exists but does not publish the dollar figure per word for any tier. [EVIDENCE NEEDED: the per word or per character price of an AI words or MT words package] | No per job minimum is stated. The floor here is the monthly subscription, not a per order minimum. |
Two notes on sourcing. The brief’s Gengo and Transifex URLs each returned an HTTP 301 redirect rather than a 200 on direct fetch, so the figures above come from the pages those redirects resolved to, both confirmed at 200. Smartling could not be included: its pricing page returned a 404 during verification, and no live pricing URL turned up through a scan of its homepage links either.
Running the math on a 40 word caption, five times over
Gengo is the only vendor here whose page confirms both a per word rate and the absence of a minimum, so it is the only one where the arithmetic can be completed. At the lowest confirmed Standard API rate, $0.03 a word, a 40 word caption comes to $1.20. At the highest confirmed Business web rate for a harder pair, $0.16 a word, the same caption comes to $6.40. Because no minimum applies, that cost scales linearly: five language pairs at the cheaper rate run five times $1.20, or $6.00 total, and five pairs at the harder rate run five times $6.40, or $32.00 total. Whether the total lands nearer $6 or $32 depends on which language pairs are involved, not on word count, since 40 words never comes close to triggering a floor that does not exist on this page.
Transifex cannot be run through the same math, because the page confirms translation volume is billed through tiered add on packages without stating what a word inside those packages costs. A 40 word caption is a rounding error against a $135 or $200 a month subscription regardless, but that is a different claim from knowing the per word price, and this piece cannot confirm the latter. DeepL Pro’s calculation is incomplete for the same reason in the other direction: no per character rate was readable on the page, so there is nothing to multiply by 40 words or by five languages. [EVIDENCE NEEDED: DeepL Pro’s per character rate, to compute what a 40 word caption costs under its tiers]
Where a machine translation API changes the arithmetic
The structural difference is between a rate with a floor underneath it and a rate without one. Gengo’s page confirms a per word rate for human translation with no minimum, so the caption’s cost is whatever the word count times the rate comes to, full stop. A machine translation API billed per character, priced the same way, would behave identically: no floor means the cost of a 40 word caption is tiny.
DeepL’s page confirms its API plans exist, including a free Developer tier, pointing toward a model that can behave this way for small jobs, but the rate itself was not readable here. [EVIDENCE NEEDED: DeepL’s per character API rate and whether it carries any monthly minimum spend]
None of this is a claim about which output reads better. A per word or per character calculation says nothing about whether a machine translated caption sounds natural in the target language or whether a human translator would phrase it differently. That question stays separate from what this piece can verify.
What to check before picking a vendor for caption-length work
- Look for “minimum,” “minimum order,” or “minimum project fee” on the pricing page itself. Gengo’s page states none of these, which is why its per word rate applies cleanly to text this short.
- Identify whether the quoted rate is for machine translation, human translation, or a hybrid before doing any math. A per word human rate and a per character machine rate are not comparable, and minimum fee logic often differs between them.
- Check whether the vendor bills each language pair separately. Gengo prices each pair on its own, turning one caption headed to five languages into five separate calculations. A subscription model like Transifex’s, where volume draws from a shared pool, does not multiply the same way.
- Re-verify the figures before budgeting against them. Rates on all three pages can change without notice, and a number confirmed on one date is not guaranteed to hold on the next.
FAQ
Why does a short caption sometimes cost as much as a longer piece of text to translate?
Because some vendors charge a minimum project fee or minimum order regardless of word count, a 40 word caption and a much longer document can land at the same price once the job falls under that threshold. Gengo’s page states no such minimum, which is why word count keeps mattering down to caption length for that vendor. Where a page does state a minimum, that minimum decides the price for text this short, not the per word rate.
Does a machine translation API avoid the minimum fee problem?
Only where the vendor’s own page confirms no minimum applies and billing is purely per character or per word. That is a fact about the pricing model, not a claim about which output reads better. DeepL Pro’s page did not yield a readable per character rate during this check, so whether it avoids a minimum in practice cannot be confirmed here.
Does the cost scale linearly with the number of languages?
For Gengo, yes: its page confirms a flat per word rate with no minimum, so five language pairs cost exactly five times what one pair costs, from $6.00 at the cheapest confirmed rate to $32.00 at the most expensive. Transifex and DeepL Pro cannot be answered either way, since neither page publishes a per word figure to test, so linear scaling is unconfirmed for both rather than assumed true or false.
Sources
- DeepL Pro pricing page, fetched 2026-09-09, HTTP 200. Confirms subscription and API plan tier names only; no readable per character rate or minimum.
- Gengo pricing and languages page (the page the brief’s pricing URL redirects to), fetched 2026-09-09, HTTP 200. Confirms per word rates by language pair and the absence of a stated minimum.
- Transifex pricing page (the page the brief’s pricing URL redirects to), fetched 2026-09-09, HTTP 200. Confirms subscription tier prices and the existence of tiered AI/MT word add on packages, without a published per word dollar figure for those add ons.
Prices verified 2026-09-09. This is not Watchdog data and this page is not on the network’s re-verification schedule.
Back to Blog