Google spotlights Gemini, Lyria, Gemma builds

This reads like marketing flywheel, not new product news: Google is trying to convert model launches into proof of real-world usage. The interesting part is the breadth of the stack, from multimodal embeddings to music generation to open edge-friendly models.

• –Gemini Embedding 2 targets multimodal retrieval and RAG, so it is aimed squarely at practical developer infrastructure rather than demos
• –Lyria 3 expands Gemini into audio generation, which broadens Google’s creative toolset beyond text and images
• –Gemma 4 keeps the open-model story credible for teams that want on-device, local, or self-hosted deployment
• –The post suggests Google wants community examples to validate these models in the wild, not just benchmark claims
• –For builders, the opportunity is cross-modal product design: search, generation, and agents built across text, image, audio, and code