Animals use previous knowledge to flexibly guide behaviour, quickly learning to respond appropriately to novel stimuli, for example using rewards as a teaching signal. Task-relevant information and behavioural states are represented globally in the brain, which suggests distributed representations during sensorimotor learning. However, these brain-wide representations have been measured in distinct expert or naive animals with ”stationary” representations. It remains unclear how de novo learning of a sensorimotor transformation involves distributed, parallel activity throughout the brain. To address this, we developed a behavioral paradigm that allows us to probe rapid reward-based sensorimotor learning in mice, overcoming the limitations of longitudinal recordings. Mice are first pretrained on an auditory detection task in which they must lick for a reward after an auditory stimulus. Once expert in this task, these mice are transferred to a whisker-based detection task, where they must also learn to lick for a reward after a novel whisker stimulus, while continuing to perform the auditory task. Mice were able to acquire the novel whisker-reward association in fast timescales, progressing from novice to high-performance levels within a single session. Furthermore, this learning manifested after only a few trials and was contingent upon the reward, as the association did not develop in a cohort of mice that did not receive a reward following whisker stimulation. During this single-session learning, we performed Neuropixels recordings using up to 4 probes simultaneously, revealing widespread task-related neural activity across the brain. We observed that whisker representations emerged in an area-specific manner and dependent on the stimulus-reward contingencies. These results indicate that reward-based sensorimotor learning recruits and modulates brain-wide representations during fast timescale learning.