Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber

Google is developing a custom server chip called Frozen v2, which integrates the architecture of Gemini directly into hardware to improve processing efficiency. Expected to launch in 2028, the chip aims to make AI inference six to ten times more efficient than current TPU technology. This shift could significantly lower Google's operational costs for running large models and provide the company with a substantial hardware advantage in the competitive AI infrastructure market.