What is Brain2Qwerty v2?

Brain2Qwerty v2 is a non-invasive brain-computer interface from Meta that decodes raw MEG brain signals into text. Using end-to-end deep learning and large language models, it achieves up to 78% word accuracy without requiring any surgical implant. The system processes magnetoencephalography (MEG) recordings from participants who type naturally, translating neural activity directly into coherent sentences in real time.

Who it's for

• Researchers in neuroscience — those studying brain decoding, language processing, or non-invasive neural interfaces who need open-source code and datasets to advance their work.
• Clinicians treating communication disorders — professionals working with patients who have brain lesions that prevent speech, seeking scalable alternatives to invasive neuroprostheses.
• AI and machine learning engineers — developers interested in applying end-to-end deep learning and LLM fine-tuning to noisy biological signals for real-time language generation.

Key features

End-to-end deep learning from raw signals

Brain2Qwerty v2 bypasses traditional hand-crafted pipelines by decoding directly from raw MEG brain recordings. This approach eliminates the need for manual feature engineering and captures neural patterns that conventional methods miss.

LLM fine-tuning on neural data

Large language models are fine-tuned on neural recordings to leverage semantic context, bridging the gap between noisy brain signals and coherent language output. This allows the system to recover full sentences even from imperfect inputs.

Real-time sentence decoding

The pipeline operates in real time, decoding brain activity into text as participants type. With a word accuracy of 61% on average and up to 78% for the best participant, it approaches performance levels previously exclusive to surgical implants.

Open-source code and dataset release

Meta is releasing the full training code for both Brain2Qwerty v1 and v2, while the Basque Center on Cognition, Brain, and Language (BCBL) releases the v1 dataset. This open approach aims to accelerate neuroscience breakthroughs across the field.

What stands out

"Brain2Qwerty v2 recovers sentences coherently from noisy neural inputs, achieving a word accuracy rate of 61%, significantly improving upon the 8% word accuracy from other non-invasive methods."

This leap from 8% to 61% average accuracy represents a fundamental shift in what non-invasive brain-computer interfaces can accomplish. The system's ability to scale performance log-linearly with data volume suggests that further gains are achievable through data collection alone, potentially closing the gap with surgical approaches entirely.

Worth checking out if…

You're working on neural decoding, communication restoration for patients with brain lesions, or open neuroscience research. Brain2Qwerty v2 is especially relevant if you need a non-invasive, scalable alternative to surgical neuroprostheses and want access to reproducible training code and datasets.