A compact but capable text embedding specialist that punches above its weight for a 400M parameter model. It focuses entirely on producing dense vector representations rather than generating text, making it single-minded and efficient. Its open weights and MIT license mean it runs anywhere without restrictions.