Product
Why we support 66 languages, not 130
· 1 min read
It is tempting to advertise a bigger number. Adding a language to a list costs nothing; the model will emit something for almost any pair. But “emits something” and “translates well” are different claims, and a list padded with the first while implying the second is a quiet lie the user only discovers when it matters.
We count what we can stand behind
Traxlate lists the languages it actually tunes and measures. Of those, a core set is the group we test hardest — measured with real quality scores on real corpora, not eyeballed. A language earns its place on the list by passing that bar, not by being grammatically possible for the model to attempt.
And the app refuses the rest
When you pick a pair we do not support, the app declines rather than producing confident nonsense. That is deliberate. A translator that guesses badly on a language it cannot do is more dangerous than one that says “not yet” — the bad guess reads fluently and you have no way to know it is wrong.
A translation you cannot verify has to be one you can trust. That trust is the whole product — so we would rather offer fewer languages we are sure of than a longer list we are not.
The list grows the honest way
New languages arrive when they clear the measurement bar, and the changelog records when. If yours is not there yet, that is a promise we have not made rather than one we are quietly breaking. You can see the current list on the Translate page.