Long documents & batch translation

Upload a whole book; it’s split into blocks and each becomes its own translation job. You don’t have to chunk it yourself, and the document’s context still travels with every block.

Translate a selection, a page, or the whole document in one go. While a batch runs you get a progress toast with the current step, blocks done, and a live line of the model’s own thinking.

Leave the document or reload the page and the batch keeps going; come back and the progress toast shows where it is.

If the connection drops, the toast shows Connection lost — reconnecting… until it is back, and asks you to refresh the page if it can’t reconnect.

Batch translation is available on every plan. A run spends from your normal credit balance; large-parallel mode needs Pro.

On documents longer than about ten pages, Customize in the translate dialog has a Translate faster with parallel agents switch in the Speed group under Advanced (1 in the figure below). With it on, the price becomes a range (2). Instead of translating front-to-back, it runs several agents at once, so a long document finishes faster.

To keep the parts consistent, the agents share a glossary and style guide that the run builds as it goes: before each group of pages, a strong (Pro) model reads them and updates the shared terms and style, then those pages translate in parallel against the same decisions. The parts can’t drift, because the key choices are made once and shared, and you can review them afterwards.

Large-parallel mode is a Pro-plan feature, offered only on long documents.

Proofreading has a parallel mode on the same terms. It reads the pages in order, each read seeing the notes from the pages before it, and writes the fixes in parallel. It costs the same as a normal proofread, with no extra fee.

Your translation bills at the model’s normal rate, exactly as a regular run. On top of that, each time the agents update the shared glossary or style guide, the pages they read for it bill 0.2 credits per source word, the same rate as building a glossary from a document. They check once for roughly every four pages, and a check that changes nothing costs nothing.

Because the exact number of changes isn’t known up front, the dialog shows the cost as a range: the low end is the translation on its own, the high end adds the reconciliation passes. The run sets aside the high end and refunds whatever it doesn’t use, so you’re only charged for what actually happens.

The shared glossary and style guide the run builds are saved as “AI sync — your document” resources, marked with an AI generated badge. When the run finishes, an “AI sync ready for review” prompt appears — click it to go straight into the review. You can also start it any time from the AI sync rows in the document’s sidebar (the Review button on the Glossary and Style tabs), or from your Glossaries and Styleguides pages.

The review itself is quick: accept or reject each glossary term, then either keep the glossary as its own or merge the accepted terms into one of your existing glossaries. For the style guide you can preview an AI-suggested merge into another style guide — the AI sync on the left, the merged result on the right — or simply save it as its own.

Until you review it, an AI sync stays attached to its own document: it won’t clutter the glossary or style-guide lists in your other documents. Once reviewed, it becomes an ordinary glossary or style guide (or folds into the one you chose), and the next parallel run on that document builds a fresh one.