Cost and speed
Small models, big results: Achieving superior intent extraction through decomposition
Our short summary of Google’s engineering write-up. The full story, with all the detail, is on their blog.
Read the full write-up on research.google (opens in a new tab)Problem
Mobile agents needed to understand user intent without sending data to servers, which is slow, expensive, and risky.
Solution
They split intent detection into two steps: a tiny multimodal model summarizes each screen, then another small model reads the summaries to predict overall intent, all on‑device.
Want the architecture, code and numbers? They’re in the original post.
Read the full write-up on research.google (opens in a new tab)