We measure the cost, time and complexity of meeting each set of quality requirements with a main solution and a backup.

★ marks a qualifying one-step solution. ☆ marks a solution that qualifies with additional steps or the gender-neutrality agent. For each language, the interactive chart shows errors per 1,000 words, cost per million source characters and time per 1,000 source words for the same qualifying solution. The chart settings select the solution with the fewest errors, lowest cost or shortest time.

For each language and level, we show one solution in the table of solutions, errors, cost and time: the simplest that meets the level, the cheapest of them where several are equally simple. The type, model, total errors, cost and time in a row are those of that solution.

In the explorer and the table of solutions, errors, cost and time, EPT means total errors per 1,000 source words across all requirements. It adds the weighted general translation quality errors (minor 1, major 5, critical 10, as in the quality bar), missed glossary terms, sentences in the wrong register, tag errors and sentences that fail gender neutrality. Each error of the other requirements counts once, and the same error types are included at every quality level.

For each requirement, we calculate the count by multiplying the share that failed by the number of terms, sentences or tags checked across the four texts. We then express the total per 1,000 source words. Formatting pass/fail results affect whether a solution qualifies but do not add to EPT.

The table of cost against errors counts only weighted general translation quality errors.

In 27 of the 78 languages with two solutions that meet every Standard requirement, the cheapest solution also has the fewest errors. Elsewhere, the solution with the fewest errors costs a median 5.1 times as much as the cheapest, for a median 4.6 fewer weighted errors per 1,000 words. In 26 languages, the fastest solution also has the fewest errors.

In 41 languages, two solutions offer different balances of cost and errors; in 9 languages, three do, and in one, four. For each of these solutions, no alternative has both a lower cost and fewer errors.