Omnilingual ASR is an open-source automatic speech recognition suite from Meta that provides native transcription for over 1,600 languages, including hundreds previously unsupported by ASR technology. It combines a family of flexible speech models (including a 7B multilingual audio representation model) with a massive speech corpus to enable scalable zero-shot learning and rapid extension to new languages using only a few paired examples. The project includes model weights, training and evaluation code, and dataset releases (via GitHub and Hugging Face), plus demo spaces for evaluation and community use. Its primary value is making high-quality speech technology accessible and extensible for low-resource and underserved language communities.