The numerical precision (number of bits) used when running a model to generate outputs; lower precision is faster but may reduce quality.