Google DeepMind just rolled out Gemma 4 12B, a 12-billion-parameter model that can parse text, images, audio, and video ...