Meta's TRIBE v2 Creates a Digital Twin of the Human Brain From fMRI Data

Meta just released TRIBE v2, a foundation model that predicts how your brain responds to almost anything you see or hear. It's a digital twin of neural activity, trained on over five hundred hours of fMRI recordings from more than seven hundred people. TRIBE stands for Trimodal Brain Encoder. It processes images, audio, and text through a three-stage pipeline that maps predictions across roughly seventy thousand brain voxels - up from around a thousand in version one. It can make zero-shot predictions for entirely new subjects, languages, and tasks without any additional training. The original TRIBE won first place at the Algonauts 2025 brain modelling competition. Version two scales that architecture dramatically and, according to Meta, often produces cleaner signals than actual fMRI scans. Meta has open-sourced the model for neuroscience, AI, and healthcare research. A digital twin of the brain that outperforms the real scanner - that's the kind of sentence that sounds like science fiction until you read the paper.