Meta open-sourced a brain-to-text system that reaches 78% word accuracy without surgery.
Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-06-30
Meta open-sourced Brain2Qwerty v2, a non-invasive brain-to-text system that achieves 61% average word accuracy and 78% for top participants by reading MEG helmet signals and correcting output with a fine-tuned LLM.
Extraction
Topics: brain-computer-interfaceneural-decodingmeta-aiopen-source-ai
Claims
- Brain2Qwerty v2 achieves 61% average word accuracy across participants and 78% for the strongest participant.
- The system uses non-invasive MEG helmet recordings rather than surgically implanted electrodes.
- More than 50% of sentences from the strongest participant contained at most one word error.
- A fine-tuned LLM repairs word and sentence errors inferred from raw brain signals, explaining large gains over prior 8% accuracy baselines.
- Performance improves as training data volume grows, suggesting further gains are achievable.
Key quotes
Brain2Qwerty v2 converts non-invasive brain recordings into text with 61% average word accuracy and 78% for its strongest participant.
More than half of sentences from the strongest participant had one word error or less.
A fine-tuned LLM then uses language context to repair likely word and sentence errors. This explains why the system beats earlier non-invasive methods reporting 8% word accuracy.