Meta builds a digital twin for the human brain
Meta just introduced TRIBE v2, a foundation model that predicts how the human brain responds to sights, sounds, and language. Trained on fMRI data from over 700 volunteers, it enables zero-shot predictions for new subjects and tasks, letting neuroscientists test brain hypotheses without recruiting human subjects for every test.
TRIBE v2 is a predictive foundation model designed to act as a digital twin of human neural activity, simulating how the brain processes sights, sounds, and language.
Instead of booking expensive scanners and recruiting volunteers for every preliminary hypothesis, researchers can feed the model text, audio, or video and watch it generate high-resolution predictions of brain responses.
How it works
The project builds upon Meta's Algonauts 2025 award-winning architecture, which was limited to low-resolution fMRI recordings from just four individuals.
For v2, the team scaled up dramatically. The training dataset incorporates fMRI brain scans from more than 700 healthy volunteers exposed to rich multimodal stimuli, including images, podcast audio, video clips, and written text.
Thanks to that scale, TRIBE v2 delivers a 70x boost in prediction resolution compared to similar models. It also unlocks true zero-shot generalization, accurately predicting neural reactions for unseen human subjects, unfamiliar languages, and novel tasks without any fine-tuning.
How to get it
Meta released the full package: the research paper, the complete codebase, and the trained model weights under a CC BY-NC license for non-commercial research.
If you want to poke around without setting up a local GPU environment, there is also an interactive web demo available to test how the model maps various media inputs onto neural activations.
Keep in mind the non-commercial license means commercial neurotech setups and clinical hardware makers cannot use the weights directly in commercial products.
Why it matters
Neuroscience has always been bottlenecked by the sheer cost of recruiting human subjects and running fMRI machines. In our view, TRIBE v2 changes that math by giving researchers a dependable dry-run engine to validate cognitive hypotheses on a laptop before spending thousands on real lab time.
While the non-commercial license keeps it out of commercial BCI pipelines for now, computational neuroscientists and academic labs should definitely grab these weights and explore what an in silico brain can do for their research workflow.
Sources (1)
- ai.meta.com(opens in a new tab)Originalai.meta.com