Mozart is a music production system for Claude. Paste it into claude.ai and type “Mozart.” Claude becomes a music studio that takes a description and returns a .wav file.
Two distinct phases. Phase 1 is composition. The model writes a symbolic score from music theory: notes, chords, time signatures, dynamics. Pure symbols, no sound. Phase 2 is execution. The model takes that score and renders it into audio using synthesis and math. The output is a .wav file. There is no audio model in the loop. No samples. No training data. Every sound is waveforms and equations.
Music notation is symbolic. Language models are good at symbols. Synthesis is math. Language models can write the math. Mozart connects the two with a strict pipeline so the model never has to hallucinate sound: it composes, then it computes.