# Meta's Non-Invasive Brain-to-Text Decoder Reaches 61 Percent Word Accuracy

Brain2Qwerty v2 reconstructs typed sentences from magnetoencephalography scans, up from roughly 8 percent for prior non-invasive methods, with code and dataset released.

- Published: 2026-07-08T05:30:03.109Z
- Canonical: https://polylog.news/ai/2026-07-08/meta-s-non-invasive-brain-to-text-decoder-reaches-61-percent
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Meta AI](https://ai.meta.com/blog/brain2qwerty-brain-ai-human-communication/), [Nature Neuroscience](https://www.nature.com/articles/s41593-026-02303-2), [MarkTechPost](https://www.marktechpost.com/2026/06/30/meta-ai-releases-brain2qwerty-v2-a-non-invasive-meg-brain-to-text-pipeline-decoding-typed-sentences-at-61-word-accuracy/)

Meta's Fundamental AI Research (FAIR) lab published [Brain2Qwerty](https://ai.meta.com/blog/brain2qwerty-brain-ai-human-communication/), a pipeline that decodes typed sentences from magnetoencephalography (MEG) recordings, with the v2 system reaching an average 61 percent word accuracy and 78 percent for its best participant. The underlying method appears in [Nature Neuroscience](https://www.nature.com/articles/s41593-026-02303-2), and per [MarkTechPost](https://www.marktechpost.com/2026/06/30/meta-ai-releases-brain2qwerty-v2-a-non-invasive-meg-brain-to-text-pipeline-decoding-typed-sentences-at-61-word-accuracy/) the training code and dataset are public.

The technical claim is a large increase for surgery-free decoding. Prior non-invasive approaches were near 8 percent word accuracy, far below the electrode implants used in clinical brain-computer interface (BCI) work. Meta trained on roughly 22,000 sentences from nine volunteers, each wearing an MEG helmet for about ten hours, and replaced hand-tuned signal-processing stages with an end-to-end deep network coupled to a language-model prior that corrects likely character sequences.

The caveats are physical, not statistical. MEG requires a magnetically shielded room and a helmet-scale sensor array, so this is a laboratory result, not a wearable device. The system also decodes typing, which produces motor signals, rather than imagined speech. What is verified is a reproducible accuracy gain on a public dataset. What is not yet shown is whether the approach transfers to portable sensors or to users who cannot move.

## What this means

The mechanism is the closing gap between non-invasive and implanted decoding, which matters to the assistive-technology and neurotech sector and to invasive-BCI companies whose advantage is fidelity. If MEG-based accuracy keeps rising and a reproducible dataset lowers the research barrier, the case for surgery narrows for some uses. The binding constraint is hardware. MEG is room-scale, so the near-term exposure is to research labs, not to consumers, until the language-model-prior approach is demonstrated on cheaper sensors such as electroencephalography (EEG).

## What to watch

- Whether the same decoding stack transfers to portable EEG at usable accuracy, which would move this from lab demonstration toward assistive deployment.
- Independent replication of the 61 percent figure on the released dataset by a group outside Meta, the standard test for a headline neuroscience result.