The company said that the model was trained on 15 trillion mixed visual and text tokens. Read More in  AI | TechCrunch