Google has announced two major advancements in its AI portfolio, spanning generative media and core infrastructure efficiency. The company introduced Lyria 3 Pro, an upgraded music generation model that enables users to create tracks up to three minutes long, significantly extending the capabilities of its predecessor. The model offers enhanced creative control, allowing users to structure compositions with elements such as intros, verses, and choruses, and is being integrated across the Gemini app, Vertex AI, AI Studio, and the Gemini API. It is also being deployed in Google Vids and ProducerAI, with access limited to paid users in some environments.
Google Research has also unveiled TurboQuant, a new AI memory compression algorithm designed to improve inference efficiency. The system reduces working memory requirements through advanced quantization techniques, potentially shrinking memory usage by at least six times without sacrificing performance. While still in the research phase, the technology targets a critical bottleneck in AI systems and could lower operational costs. The findings are set to be presented at ICLR 2026, marking a significant step toward more efficient AI infrastructure.
