A model that has already completed its broad general learning. It arrives knowing language and a great deal about the world, ready to be adapted rather than built from nothing. This is why smaller organisations can build capable AI products without ever training a model from scratch, which almost none of them could afford.
Technical view
This is transfer learning: representations learned on a general task carry over usefully to specific ones. Because knowledge is frozen at the training data cutoff, models have no awareness of later events unless given access to search or supplied with current information in the prompt.
Related
Part of AI Terminology Index, section: Building and Using Models.