Meta Launches Brain2Qwerty v2 for Non-Invasive Brain-to-Text Translation

Meta has introduced Brain2Qwerty v2, a non-invasive AI system that translates brain activity into text with 61% average word accuracy.

Brain2Qwerty v2 records neural signals via MEG scanner. The helmet-like magnetoencephalography device captures raw brain activity while participants type. An end-to-end deep learning model then reconstructs intended sentences, with large language models fine-tuned on neural data to leverage semantic context for noisy signals.

Meta trained the system on approximately 22,000 sentences from nine volunteers, each recorded for 10 hours. The company avoided hand-crafted pipelines, relying instead on direct decoding from raw signals.

Accuracy jumps from 8% to 61% for non-invasive methods. Brain2Qwerty v2 reaches levels previously seen only with invasive surgical implants. Meta reports that decoding performance improves as training data volume grows, with AI agents used to explore pipeline optimizations before final selection.

Code, dataset, and $5 million fund support open research. Meta released training code for Brain2Qwerty v1 and v2 as part of its Digital Brain Project. Its research partner is releasing the v1 dataset, while the project includes a $5 million fund for open neuroscience datasets.

The work targets people who lost communication ability due to brain lesions. In an accompanying Nature Neuroscience paper, Meta researchers noted that high-performing brain-computer interfaces have largely required implanted electrodes, citing surgery risks and long-term maintenance challenges.

Non-invasive approach aims to scale beyond implants. Meta positions Brain2Qwerty v2 as a bridge between invasive neuroprosthetics and non-surgical systems. The company hopes open collaboration will accelerate identification, diagnosis, and treatment of neurological disorders.