Meta AI Presented Brain2Qwerty v2 — Non-invasive MEG Text Decoder with 61% Accuracy

Meta AI on June 30, 2026 presented Brain2Qwerty v2 — a non-invasive (without surgical implants) brain-to-text pipeline based on magnetoencephalography (MEG), which decodes printed sentences with 61% word accuracy, and published open source code for training the model.

What is Brain2Qwerty v2

• Developer — Meta AI research division
• Method of reading brain signals — magnetoencephalography (MEG), without implants
• Declared decoding accuracy — 61% of words guessed correctly (word accuracy)
• The model's task — to restore printed sentences from brain signals
• Code for training the model published in open access

How Does a Non-Invasive Brain-Computer Interface Work?

The MEG helmet captures weak magnetic fields that arise from electrical activity in brain neurons — without surgery and without implants, unlike invasive brain-computer interfaces (BCI), where electrodes are placed surgically directly on or under the brain cortex. The name Brain2Qwerty refers to the QWERTY keyboard layout: the pipeline translates the captured signal into letters and words corresponding to what a person types.

Non-invasive methods like MEG typically produce a more "noisy" and less accurate signal compared to implants, but do not require brain surgery — this makes them potentially more scalable for research and, in perspective, for a wider range of users.

The problem formulation itself — decoding specifically typed text, rather than, say, spoken or imagined speech — is important because typing on a keyboard provides researchers with a clear, labeled signal: when a person presses a specific key, this action can be precisely matched with MEG helmet readings at the same moment in time. Such labeling simplifies model training compared to more vague tasks like decoding internal speech.

Why Does Open Source Code Matter?

Meta AI published not only results but also model training code — this allows other research groups to reproduce the experiment, verify the claimed 61% word accuracy, and develop their own pipeline variants based on the same architecture. For a relatively young field of non-invasive BCI, code openness accelerates the accumulation of comparable results between different laboratories.

What It Means

Brain2Qwerty v2 is another step in the development of non-invasive brain-computer interfaces: 61% word accuracy is still far from invasive solutions, but confirms that text decoding from brain signals without surgery is gradually becoming a measurably workable task, not just a laboratory hypothesis.