Meta Unveils NLLB-200: Open‑Source AI Model Translating 200 Languages
Meta has open‑sourced its new NLLB‑200 model, a single AI system that translates 200 languages with up to 44 % higher quality than its predecessor, supporting numerous low‑resource languages and powering billions of daily translations across Facebook and Instagram to improve user experience and content safety.
Meta (Facebook) announced the completion of a single AI model, NLLB‑200, capable of translating 200 languages, and will open‑source both the model and its training dataset.
The model is part of Meta’s high‑quality machine‑translation system for the metaverse, improving translation quality by an average of 44 % over its predecessor.
It follows the 2020 open‑source release of the M2M‑100 model, which could translate 100 languages without English as an intermediate.
Meta highlights NLLB‑200’s support for low‑resource languages such as Kamba and Lao, and notes that it covers 55 African languages compared with fewer than 25 supported by many commercial services.
In benchmark FLORES‑101 tests, NLLB‑200 achieved BLEU scores 44 % higher on average, with up to 70 % gains for certain African and Indian languages.
The system will power up to 250 billion daily translations across Facebook’s News Feed, Instagram and other platforms, enhancing user experience and helping detect harmful content and misinformation.
Editor: Da Xiong
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
21CTO
21CTO (21CTO.com) offers developers community, training, and services, making it your go‑to learning and service platform.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.