MUM was announced by Pandu Nayak, Google Fellow and Vice President of Search, on 18th May 2021. The name stands for "Multitask Unified Model".
Google's claims were specific. MUM "uses the T5 text-to-text framework and is 1,000 times more powerful than BERT", and "it's trained across 75 different languages and many different tasks at once".
It was announced as a capability, not a ranking update. Google said: "We'll bring MUM-powered features and improvements to our products in the coming months and years."
Google's ranking systems guide now lists MUM among its active systems. It describes an AI system for understanding and generating language, used for specific tasks rather than general ranking.
MUM followed BERT, the 2019 language model Google applied to ranking and featured snippets.