Introducing TRIBE v2: A Predictive Foundation Model Trained to Understand How the Human Brain Processes Complex Stimuli
Meta AI
Meta built an AI that predicts how your brain reacts to images, sound, and text, using data from 700+ people. It's a digital stand-in for real brain scans, and they're giving it away for research.
Based on reporting by Meta AI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Meta's neuroscience team just dropped TRIBE v2, a foundation model trained to forecast human brain activity in response to whatever you throw at it: a photo, a podcast, a video clip, a paragraph of text. It's a sequel to the model that won Meta an award at Algonauts 2025, but the scale-up is the real story here. That earlier version leaned on low-resolution fMRI data from just four people. TRIBE v2 was trained on high-resolution scans from more than 700 healthy volunteers, and that jump in sample size is what lets it generalize.
And generalize it does. The model can make zero-shot predictions for brains it's never seen, languages it wasn't trained on, and tasks outside its original training set. That's a meaningful technical leap for a field where most models are built around a handful of subjects and fall apart the moment you ask them to work on someone new. Consistently beating standard modeling approaches on this kind of prediction task isn't a small claim in neuroscience circles.
The pitch from Meta is practical: build a good enough digital proxy of the brain, and researchers stop needing a live human in a scanner for every single hypothesis they want to test. Running experiments on a model instead of recruiting subjects, scheduling scan time, and waiting weeks for results could compress research timelines dramatically. That matters for basic science, but it also matters for clinical work, where understanding how injured or atypical brains process stimuli could eventually shape diagnostics or treatment planning.
Meta is releasing the research paper, model weights, and code under a CC BY-NC license, plus a demo site so people can poke at it directly rather than take the claims on faith. Non-commercial licensing keeps it out of product pipelines for now, but it's open enough for academic labs to build on immediately. That's the more interesting bet: not that Meta will use this internally, but that handing it to neuroscientists worldwide compounds faster than keeping it locked up.
My take — AI-written commentary, not fact-checked reporting
I like seeing a big lab put out a genuinely open research artifact instead of another chatbot wrapper, and neuroscience is exactly the kind of field that benefits from shared infrastructure rather than everyone rebuilding brain models from scratch. The non-commercial license is the right call here too — this should live in labs, not get quietly folded into an ad-targeting pipeline six months from now.
Read more about this at: Meta AI