Meta AI unveils Brain2Qwerty v2, a noninvasive MEG text decoder with 61% accuracy

On June 30, 2026, Meta AI unveiled Brain2Qwerty v2, a noninvasive brain-to-text pipeline that requires no surgical implants. It uses magnetoencephalography (MEG) to decode typed sentences with 61% word accuracy. Meta also published open-source code for training the model.

What is Brain2Qwerty v2

• Developer: Meta AI's research division
• Brain signal capture: magnetoencephalography (MEG), no implants
• Claimed decoding accuracy: 61% of words correctly predicted (word accuracy)
• Task: reconstruct typed sentences from brain signals
• The model's training code is publicly available

How does a noninvasive brain-computer interface work?

An MEG helmet picks up the faint magnetic fields generated by the electrical activity of neurons in the brain. It needs no surgery and no implants. Invasive brain-computer interfaces (BCIs) are different: surgeons place their electrodes directly on or beneath the cortex. The name Brain2Qwerty refers to the QWERTY keyboard layout. The pipeline converts the recorded signal into the letters and words a person is typing.

Noninvasive methods like MEG typically produce a noisier, less precise signal than implants. However, they don't require brain surgery. That makes them potentially more scalable for research and, eventually, for a broader range of users.

The choice of task matters. The model decodes typed text, rather than spoken or imagined speech. Typing gives researchers a clean, labeled signal: when a person presses a specific key, researchers can precisely match that action to the MEG helmet's readings at the same moment. This labeling makes the model easier to train than fuzzier tasks such as decoding inner speech.

Why does the open code matter?

Meta AI published both its results and the model's training code. Other research groups can now reproduce the experiment, verify the claimed 61% accuracy, and build their own variants of the pipeline on the same architecture. Noninvasive BCI is a relatively young field, and open code helps labs accumulate comparable results faster.

What this means

Brain2Qwerty v2 is another step forward for noninvasive brain-computer interfaces. Its 61% word accuracy still falls well short of invasive systems. However, it shows that decoding text from brain signals without surgery is gradually becoming a measurable, workable task rather than just a lab hypothesis.