GlucoFM
A very small Google Research model that interprets continuous glucose monitor data to spot metabolic patterns.
GlucoFM is a research model from Google published in August 2026 that reads traces from continuous glucose monitors, the sensors worn under the skin that measure blood sugar every few minutes. Its trick is separating the slow drift of a person’s baseline from the sharp short spikes caused by meals or exercise, then combining the two, rather than treating the whole trace as one undifferentiated signal.
The striking part is the size: roughly 720,000 parameters, which is around a millionth of a large chat model and small enough to run on a phone. It was pre-trained on more than 109,000 hours of unlabelled sensor data so that it only needs a handful of expensive labelled clinical examples to be useful for a specific task. It is research, not a product or a medical device.