Meta Unveils Noninvasive Brain2Qwerty v2 With 61% Word Accuracy

Brain-computer interfaces aim to restore communication for people who have lost speech or motor function because of brain injuries, amyotrophic lateral sclerosis or other conditions. Most high-performing systems require implanted electrodes, bringing surgical and long-term care risks. Meta AI and the Basque Center on Cognition, Brain and Language, or BCBL, instead used magnetoencephalography, or MEG, to measure brain signals from outside the scalp while participants typed. The work remains a study involving healthy participants.

Meta unveiled Brain2Qwerty v2 on June 29, 2026. Nine volunteers each typed for 10 hours while wearing an MEG device, generating a total of about 22,000 sentences. An end-to-end deep-learning system combined with a fine-tuned LLM achieved average word accuracy of 61%, rising to 78% for the best-performing participant. Meta has released the training code for v1 and v2, while BCBL has published the v1 dataset.