Breaking

Meta Releases Brain2Qwerty v2, Turning Brain Activity Into Text Without Surgery

June 29, 2026 at 18:01 EDT

Foundation Models
Research & Papers
Open Source

On June 29, 2026, Meta released Brain2Qwerty v2, a non-invasive AI system that generates text in real time from brain activity recorded by a head-worn device. It requires no surgery or brain implant, and the company has published the code and data for the research community. According to Meta AI's announcement, v2 is described as the best-performing end-to-end pipeline that decodes magnetoencephalography (MEG) signals into sentences directly from raw data.

June 29, 2026 · Meta AI

Brain2Qwerty v2 turns brain activity into full sentences

A non-invasive deep-learning pipeline decodes sentence-level text from raw brain signals recorded mainly via MEG — reaching 61% average word accuracy across nine volunteers, with a peak of 78%.

61%

Average word accuracy across 9 volunteers

78%

Peak word accuracy, best participant

22k

Sentences · 10 hrs MEG recording per person

FROM v1 TO v2 — accuracy gains

v1 measured character error rate (lower is better); v2 measures word accuracy (higher is better). Different metrics — shown to scale within each.

v1 · MEG character error rate

32%

avg error

19%

best error

v2 · word accuracy

61%

avg accuracy

78%

best accuracy

THE DECODING PIPELINE

CNN

features over 500ms windows

→

Transformer

sentence-level output

→

LLM fine-tuning

coherent error correction

The promise

Safe and non-invasive, unlike surgical implants. Code for v1/v2 is open-sourced, aiming to support people who cannot speak due to brain injury and other conditions.

The reality check

MEG hardware weighs ~500kg, costs ~$2M, needs a shielded room, and the subject must stay still. It decodes intended keystrokes — not "thoughts."

What still stands between this and real-world use

Miniaturization

shrinking 500kg, $2M hardware

Real-time performance

EEG accuracy still ~40%

Diverse validation

tested on just 9 volunteers

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.