CLaMP 3 is an advanced music information retrieval model that aligns features of musical scores, performance signals, audio recordings, and multilingual text through contrastive learning, supporting cross-modal and cross-lingual music retrieval. It demonstrates strong generalization capabilities by handling unaligned modalities and unseen languages. Trained on the large-scale dataset M4-RAG, which covers various global music traditions, the model supports a variety of music retrieval tasks such as text-to-music and image-to-music.