We introduce GAPS (Guitar-Aligned Performance Scores), a new dataset of classical guitar performances, and a benchmark guitar transcription model that achieves state-of-the-art performance on GuitarSet in both supervised and zero-shot settings. GAPS is the largest dataset of real guitar audio, containing 14 hours of freely available audio-score aligned pairs, recorded in diverse conditions by over 200 performers, together with high-resolution note-level MIDI alignments. These enable us to train a state-of-the-art model for automatic transcription of solo guitar recordings which can generalise well to real world audio that is unseen during training. In addition, we propose a novel postprocessing step to estimate note velocities in the absence of ground truth training data. For each track in the dataset, we provide metadata of the composer and performer, giving dates, nationality, gender and links to IMSLP or Wikipedia. We also analyse guitar-specific features of the dataset, such as the distribution of fret-string combinations and alternate tunings. This dataset has applications to various MIR tasks, including automatic music transcription, score following, performance analysis, generative music modelling and the study of expressive performance timing.