Tripo AI 3D Generator
Mini toolsBrain2Qwerty v2
Brain2Qwerty v2 is a non-invasive brain-to-text decoding model that processes raw MEG brain signals and outputs coherent text in real time, without requiring any surgical procedure. The system uses end-to-end deep learning trained on roughly 22,000 sentences recorded from nine participants, each wearing an MEG device for 10 hours while actively typing. Rather than relying on hand-crafted neural event detection, the model decodes directly from raw signals and fine-tunes large language models on neural data to leverage semantic context, compensating for the inherent noisiness of non-invasive recordings. Brain2Qwerty v2 achieves 61% word accuracy overall and 78% for the best participant, where more than half of sentences are decoded with one word error or fewer. Accuracy scales log-linearly with data volume. The full training code for v1 and v2 is publicly released, along with the v1 dataset via partner BCBL. Intended for assistive communication for individuals with brain lesions or conditions preventing speech or typing.
New
Multimodal
Gen 3
Overview
End-to-end deep learning pipeline that decodes text in real time from non-invasive magnetoencephalography (MEG) brain recordings, requiring no surgical implant. Trained on approximately 22,000 sentences from nine participants, it achieves 61% word accuracy -- far above the 8% benchmark for other non-invasive methods -- by fine-tuning large language models on raw neural signals.
About Meta Platforms
We're connecting people to what they care about, powering new, meaningful experiences, and advancing the state-of-the-art through open research and accessible tooling.
Industry:
Technology, Information and Media
Company Size:
78865
Location:
Menlo Park, California, US
Website:
ai.meta.com
Last updated: June 30, 2026
