omg finally someone's working on multimodal models that aren't utter garbage, can't wait to dig in and see if this is actually the real deal or just more ai hype https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/