How AI translation quality checks are priced

Automated quality checks are typically metered rather than unlimited, and metering is usually expressed in credits. That is sensible — the work has a real compute cost — but it can make the price feel opaque. Here is how to reason about it.

What a credit represents

A credit is a unit of AI work. A quality check consumes credits in proportion to the size of what it processes — a short document costs less than a long one — because the system is re-translating the source and comparing it against the delivery, which is real work, not a lookup.

What you get to start

A free allowance lets you calibrate before you commit. Glosseum includes three thousand AI credits a month on its free tier, which is enough to run genuine checks on real jobs and see, on your own documents, what the tool catches and what it costs.

The honest limits

• Automated QA is strongest on factual discrepancies — numbers, dates, names — and on estimating how machine-produced a translation reads.
• It is a gate and a second pair of eyes, not a substitute for a reviewer’s judgement on tone, register, and meaning.
• A credit model rewards checking what matters most; it is not designed for re-running everything endlessly for its own sake.