On July 16, 2026, Zyphra introduced ZUNA1.1, a groundbreaking EEG foundation model licensed under Apache 2.0. This model is designed to enhance the analysis of scalp-EEG data by utilizing a 380M masked diffusion autoencoder.
ZUNA1.1 stands out for its ability to reconstruct, denoise, and upsample EEG signals across various channel layouts. This flexibility allows researchers and practitioners to work with input lengths ranging from as short as 0.5 seconds to as long as 30 seconds.
The release of ZUNA1.1 marks a significant advancement in the field of EEG analysis, providing a powerful tool for those looking to leverage machine learning techniques in neuroscience.