Meta Releases Brain2Qwerty v2: Non-Invasive Brain-Computer Interface Achieves 61% Decoding Accuracy

On June 29, 2026, Meta released Brain2Qwerty v2, a non-invasive brain-computer interface system that converts brain activity into full sentences in real time without surgery. The system uses magnetoencephalography (MEG) signals and an end-to-end deep learning architecture combining Conformer, aligner, and a large language model (LLM) for three-level decoding from characters to sentences.

Technical Breakthroughs

• Decoding Performance: Average word accuracy of 61% across 9 subjects, with the best subject reaching 78%. Over half of the sentences had only one word or fewer errors. This is a significant improvement over traditional non-invasive methods (~8% accuracy).
• Architecture Upgrade: v2 upgrades from character-level decoding in v1 to word- and semantic-level decoding, outputting complete sentences directly. The model consists of a Brain Encoder (Conformer module) and NeuroLLM (LLM based on Qwen3-4B), aligning neural features with word embeddings via contrastive loss.
• Training Data: Each subject wore MEG equipment for about 10 hours, collecting approximately 22,000 sentence samples of imagined typing. Research confirms that decoding accuracy increases logarithmically with data volume.

Comparison with Invasive Approaches

Invasive BCIs (e.g., Neuralink) achieve over 90% decoding accuracy but require craniotomy, posing high risks and limiting accessibility. Brain2Qwerty v2 approaches invasive performance non-invasively, laying the foundation for large-scale applications.

Open Source and Ecosystem

Meta fully open-sourced the training code for v1 and v2, and partner BCBL simultaneously released the v1 dataset. Additionally, Meta released the perceptual encoding Tribev2, brain data processing tool NeuralSet, evaluation platform NeuralBench, and invested $5 million in open brain science datasets.

Challenges and Outlook

• Insufficient Accuracy: Current models still have word or character errors, making them unsuitable for daily communication.
• Equipment Limitations: MEG systems are bulky and expensive (millions of dollars), relying on liquid helium cooling and magnetic shielding. Next-generation portable OPM-MEG is still under development.

The technology aims to serve patients with communication disorders such as stroke and ALS, enabling communication without surgery. In the long term, it may drive changes in the diagnosis and treatment of neurological diseases.