Sabi's EEG-packed "Brain Foundation" beanie claims 30-words-per-minute thought-to-text — but no evidence yet
Palo Alto startup Sabi is developing a beanie embedded with 100,000 EEG sensors that it claims can translate brain signals into digital text at 30 words per minute, powered by a proprietary AI model called "Brain Foundation" trained on 100,000 hours of data from roughly 100 volunteers. The pitch is a non-surgical alternative to implanted BCIs like Neuralink — a wearable that could theoretically give mass-market access to brain-computer interfaces.
Experts are skeptical. A peer-reviewed paper in Scientific Reports found EEG-to-text model performance is "unclear due to limitations in evaluation methodologies," with early promise likely reflecting pattern memorization rather than genuine brain-wave decoding. Third-party neurotech consultant JoJo Platt noted the device will need to work out of the box without user adaptation — a bar Sabi has yet to demonstrate publicly. The product is slated for a 2026 release with no independent benchmarks shared so far.
Sources
- Primary source
- Wired — original reporting on Sabi's Brain Foundation model training data
“The AI model powering it is said to be trained on 100,000 hours of data from some 100 volunteers”