DeepL

What DeepL actually meters

DeepL meters a character and defines it properly. One character is one Unicode code point of the source text, so "A", "Δ", "あ" and "深" each count once. Spaces, tabs and line breaks count. Characters inside HTML or XML tags don't when tag handling is on, and neither does the API's context parameter.

Then it meters a second thing that isn't characters. Translator plans cap file translations by count, not size: 3 a month on Individual, 20 per user on Team, 100 per user on Business. A 40-page contract and a one-line memo each spend one.

The API inverts that. There, documents draw on the same character balance, and every .docx, .doc, .pptx, .xlsx or .pdf bills a minimum of 50,000 characters however short it is. At €22.00 per million, that's a floor of €1.10 per file. HTML, SRT, TXT and XLIFF bill to the exact character.

Voice is a third unit. Paid Translator seats include 30 minutes of DeepL Voice per user per month, and the API sells audio by the hour: €1.96 per speech-to-text hour and €3.93 per speech-to-speech hour.

How the character allowance works

The allowance attaches to the seat, not the account. DeepL's plan comparison labels every column "characters (per user/month)", so a five-seat Team subscription buys five separate million-character buckets, not one shared pool.

That's the opposite of how the same page treats files. File translations default to per-user limits, but an admin can make the monthly total freely accessible to everyone, or set a fixed per-member cap that has to be identical for all members. Admins stay unlimited. DeepL wrote a help article explaining that for files and nothing equivalent for characters.

Rollover isn't documented either way. Nothing says an unused Team allowance survives into the next cycle, and the Voice allowance, which DeepL does document in detail, resets monthly with no carry-over and no pooling. That's the closest published analogue.

Business and Enterprise replace the number with "unlimited" under a footnoted fair usage policy whose threshold isn't published.

What happens when you hit the limit

On the API, the behaviour is precise: requests fail with HTTP 456, quota exceeded, and the docs tell clients never to retry it. Cost control lets a Pro, Growth or Enterprise subscriber cap monthly characters, speech-to-text minutes or speech-to-speech minutes, and DeepL stops processing there until the period ends. Key-level limits sit underneath, and raising a limit sends a notification email as a takeover check.

On Translator seats, it isn't documented. A full-text sweep of DeepL's 264 English help articles returned per-request caps for the Gmail and Office add-ins, the API's 456 error and the Voice rules, and nothing on character 1,000,001.