Three things decide what you pay, and there is no per-seat licence or minimum commitment. First, your volume: Quality Estimation is charged per character, counting source and target, while Automated Post-Editing is charged per source word of the segments it edits — segments that pass your threshold are never billed. Second, your rate band: your rate is determined by your total volume and your processing method (batch or real-time). Third, your billing model: buy prepaid credits up front, or pay monthly for what you actually used. Custom quality estimation models are quoted separately.

It differs between the two. Quality Estimation is charged per character and reads both source and target translation. Automated Post-Editing is charged per source word of the segments it edits; segments that pass your threshold are not billed.

No. API error responses do not consume volume. Similarly, if the system cannot provide an Automated Post-Editing suggestion for a segment, this does not count against your balance.

The Sandbox API has specific limits. For production, the default key has standard per-minute request and character limits, as documented in the API reference documents. Higher throughput is available on request. Standard API rate limitations apply to all usage, including uploads through the EPIC app.

See the developer docs → .

Not long. A trial API key is issued immediately, and if you're using the EPIC app or one of the existing integrations you can be scoring content the same day — there's nothing to install. Building your own connection is a small piece of work against a documented REST API; most teams have it running inside a sprint. Custom quality estimation models are the exception: those need your data and a build cycle, so we agree a timeline with you up front.

Once your trial balance is exhausted, you will need to top up your balance or contact us to discuss a post-paid plan. We may manually disable your API key or reach out to discuss your requirements.

No. Seats and reviewers are unlimited. You pay for the content you process through the API, not for the people reading the results.

At least 25,000 reviewed segments per language pair and domain, plus your glossaries and style rules. If you don't have that, TAUS can build the dataset for you.

Yes. Quality Estimation identifies the machine-translated content that already meets your quality threshold so it can skip human review entirely, and Automated Post-Editing fixes a large share of what remains before a linguist opens the file. Together these typically remove the majority of a post-editing budget.

EPIC prices Quality Estimation on characters rather than words, because a word is a different length in every language. As a rough conversion, a word averages around six characters, and QE reads both the source and the translation — so a translated word is roughly twelve billed characters. Your rate depends on your total volume and on batch or real-time processing, so ask us for a quote against your own numbers.

In every deployment we have measured, yes — because the point is not to replace review but to stop paying for review that changes nothing. Customers report reductions of up to 80% on cost and up to 76% on lead time, and Uber saved more than $500,000 in its first year by scoring support content instead of staffing multilingual reviews across 50+ languages.

Read the case studies →

No. EPIC's Quality Estimation is reference-free: it scores a translation by reading the source and the target together, with no human translation to compare against. That is what makes it usable on live content in production rather than only on test sets.