Zyphra Releases ZUNA - BCI Foundation Model Advancing Towards Thought-to-Text

New brain-computer interface AI model improves real-world EEG data while advancing Zyphra's mission to develop human-aligned superintelligence

SAN FRANCISCO, Feb. 18, 2026 /PRNewswire/ -- Zyphra today announced the release of ZUNA, the company's first foundation model trained on brain data. ZUNA significantly improves the quality and usability of electroencephalography (EEG) data while establishing an early technical foundation for thought-to-text, the direct communication between human thought and AI systems enabled by brain–computer interfaces (BCIs).

ZUNA is a 380M-parameter diffusion autoencoder model that delivers immediate value for EEG practitioners. While scalp EEG data is widely available and noninvasive, it's often messy and incomplete. ZUNA reconstructs high-fidelity brain signals from imperfect data, improving diagnostics, research workflows, and BCI applications. It also predicts missing channels from sparse inputs and electrode coordinates, scaling seamlessly from consumer headsets to 256-electrode research systems.

"We believe the next major modality in AI beyond language, audio, and vision will be thought-to-text enabled by noninvasive BCIs," said Paul White, Chief Business Officer of Zyphra. "ZUNA is a step toward that vision, and it solves everyday problems EEG practitioners face today. We are releasing ZUNA open source and want to collaborate with the community to deliver value today while continuing to innovate."

ZUNA is designed to work across a wide range of EEG systems, from fewer-channel consumer headsets to high-density clinical equipment. It adapts to different sensor layouts and recording conditions. This flexibility allows it to be easily deployed across industries including medical devices, neuroscience research, digital health, and consumer neurotechnology.

ZUNA learned the shared structure of brain signals across a wide range of devices, sensor layouts, and recording conditions using deep learning techniques applied to a diverse set of real-world EEG data.  This approach allows ZUNA to generalize beyond any specific electrode configuration, even with incomplete or noisy data. As a result, ZUNA consistently outperforms spherical-spline interpolation, the industry-standard method implemented in MNE, particularly at higher scaling factors where traditional interpolation begins to break down.

ZUNA's capabilities and performance make it a valuable tool for EEG practitioners today, while delivering a strong foundation for future thought-to-text models that will interpret and decode human thoughts via noninvasive BCIs.

Availability

ZUNA is released as open-source software under a permissive Apache 2.0 license, enabling immediate adoption and integration by researchers, clinicians, and organizations worldwide.

• Model weights: available on Hugging Face
• Inference and preprocessing code: available on GitHub
• Pip install Python package: pip install zuna

Organizations and researchers interested in collaborating with Zyphra to improve future versions of ZUNA for specific use cases are encouraged to reach out to [email protected].

For more information please reference the technical paper and the Zyphra blog post or visit www.zyphra.com.

About Zyphra

Zyphra is an open-source superintelligence company based in San Francisco, CA on a mission to build human-aligned AI that helps individuals and organizations reach their fullest potential.

Media Contact:
Paul White
Chief Business Officer
[email protected]
www.zyphra.com

SOURCE Zyphra