Abstract: Developments in deep learning methods have advanced the idea of generalizable models of neural processing, sometimes called foundation models. In the auditory system, a generalizable model should be able to explain the activity of neurons in response to arbitrary sounds in any behavioral context. We collected single-neuron activity from primary auditory cortex of awake ferrets during presentation of a large natural sound library and trained a convolutional neural network (CNN) to predict time-varying neural activity. The CNN substantially outperformed existing coding models and was able to generalize to novel stimuli, including synthetic sounds outside of the training set. However, investigation of how non-auditory context influenced activity revealed a more complex story. Traditional models implicate attention and motor activity as the primary modulators of AC, but recordings in free-moving, behaving animals indicated a dominant effect of head position and angle. These results argue that generalizable context-dependent models of sensory processing will require revising existing frameworks for behavioral context.