Meta’s Omnilingual Automatic Speech Recognition (ASR) system was included in TIME’s list of the Best Inventions of 2026. The research team from Meta’s Fundamental AI Research (FAIR) division originally released the tool as open source on November 10, 2025.
The system supports over 1,600 languages, including approximately 500 that previously lacked ASR support. Meta released a family of models with sizes ranging from 300 million to 7 billion parameters, utilizing a zero-shot adaptation approach for new languages.
The project includes the Omnilingual ASR Corpus, a dataset covering 350 underserved languages. Meta distributed the technology under an Apache 2.0 license, which permits both commercial and research applications.
Regarding performance, Meta reports a character error rate of less than 10% for 78% of the supported languages. The remaining 22% of languages currently exceed this error threshold.
