Researchers at Meta’s Fundamental AI Research team have developed TRIBE v2, a digital twin model capable of predicting human brain activity by processing integrated visual, auditory, and language stimuli.

Key points

• The TRIBE v2 model utilizes a foundation architecture to predict high-resolution brain responses across diverse tasks and subjects.
• Researchers trained the model using a unified dataset containing over 1,000 hours of fMRI data from 720 individual subjects.
• The project builds upon the original TRIBE model, which previously secured first place in the 2025 Algonauts Challenge for computational neuroscience.
• This digital twin approach significantly outperforms traditional linear encoding models in accuracy when analyzing complex, multisensory human brain activity.

Why it matters: This development provides a scalable, in silico alternative to invasive human brain research, potentially accelerating breakthroughs in diagnostics and precision medicine. As the neurology AI market approaches a projected $2.5 billion valuation by 2030, such foundation models offer a more robust framework for understanding complex neurological functions.