Meta Open-Sources TRIBE v2 AI Model to Simulate Human Brain Activity

Meta’s Fundamental AI Research, or FAIR, developed TRIBE v2, a single multimodal model that predicts fMRI responses as people view images, listen to sounds and process language. The model can run virtual experiments without rescanning participants, potentially reducing neuroscience research’s reliance on expensive equipment and large pools of subjects.

Meta released TRIBE v2’s code, weights and interactive demo on March 26, 2026. Its training data covered 720 people and more than 1,000 hours of fMRI recordings, while its prediction scope expanded from about 1,000 brain regions in the previous version to roughly 70,000 voxels. Accuracy improved by about two to three times across multiple tests, and some group-level predictions came closer to the average response than a single real-world scan did.