Mansa: multilingual AI for 30+ African languages, now live
African Languages Lab has launched Mansa, a multilingual, multimodal AI system built for African languages, accents and contexts. It is available now on web and mobile, with an API for developers and businesses.
African languages account for less than 1% of the data used to train modern AI. In Common Crawl, one of the industry's largest training sources, English makes up over 40%, while all 24 African languages tracked, including Swahili, Hausa, Yoruba and Zulu, combine to less than 0.05%. The result is familiar to anyone working on the continent: models that trip over African accents, misread idioms and lose cultural context.
Mansa: multilingual AI for 30+ African languages, now live
Mansa runs on more than 100 billion curated tokens and over 13,000 hours of expert-validated speech, gathered through open licensed sources, direct community partnerships, and All Voices, our platform where native speakers contribute and validate data and are paid for it. Thirty languages are in production, drawn from a collection spanning more than 70.
Six product lines are live at launch: Mansa Chat, Mansa Agent, Mansa Translate, Mansa Transcribe, Mansa TTS and Mansa Interpret, alongside the API.
Schools and health organisations across the continent receive Mansa at no cost.