Meta’s No Language Left Behind project reported that its flagship model, NLLB-200, improved translation quality by an average of 44% over the previous state-of-the-art systems, measured by the BLEU metric on the FLORES-200 benchmark. NLLB-200 translates directly among 200 languages and was deliberately weighted toward low-resource languages, containing three times as many low-resource as high-resource languages. The 44% figure spans the full set of languages the model covers, including many that had no high-quality automated translation before.