Meta's Non-Invasive Brain Decoder Reaches 78% Sentence Accuracy

Meta AI's Brain2Qwerty v2 decoded continuous natural sentences from magnetoencephalography (MEG) recordings of healthy volunteers with up to 78% word accuracy as of June 29, 2026. The system uses deep learning to translate brain signals into text without surgical implants. The advance moves beyond isolated word decoding to sentence-level output, where phonological and syntactic context propagates across the signal in real time.