Developed by Meta Superintelligence Labs from the ground up in just nine months, this first model is natively multimodal across text, images, and voice. Its “Contemplating” mode runs multiple AI agents in parallel and powers Meta AI across Facebook, Instagram, and WhatsApp.