Simple, Transparent Pricing

Every untranslated row can be sent to one or more machine translation engines (Amazon, Azure, DeepL, Google, and others, depending on what's configured) for an instant first-draft translation, right from the translation grid or automatically when a project is built.

You can check several engines at once to compare their results and pick the best one, or to gain confidence when they agree - each engine checked counts separately against the plan's machine translation quota. Usage is billed per word of the source text, per engine queried, so translating the same text with two engines costs twice what translating it with one engine would.

This is a capacity limit, not a usage quota that resets each month - it's the total number of source words your account's projects can hold at once. A project's word count is counted once, plus a further 25% of that count for every target language you've added to it, so a 1,000-word project with 3 target languages counts as 1,750 words toward this limit (1,000, plus 25% of 1,000 x 3).

Going over the limit doesn't affect projects you already have - it only stops you from adding or updating source content (new files, new languages) until you either free up capacity elsewhere or move to a plan with a higher word count.

This counts everyone with standing access to your account - the owner, plus everyone you add from Users on the project page: account admins, project managers, translators, and anyone else you invite to help manage or translate your projects.

Every translation can be checked by an AI model (e.g. Claude) that compares it against the original text and flags anything that looks wrong - a mistranslation, a dropped or altered placeholder like %s or {0}, mismatched punctuation, that kind of thing. When it finds an issue, it suggests a fix you can apply with one click, right from the translation grid.

You can run more than one AI model at once for a second opinion - each one checked counts separately against the plan's validation quota. Usage is billed per word of the original and translated text combined, per model queried, so a longer string checked with two models costs twice what checking it with one model would.