Quantization is a technique used to make AI models, especially large language models, run more efficiently. It involves reducing the precision of the numbers used to represent a model's weights and activations—for example, storing them as 4-bit or 8-bit integers instead of 32-bit floating-point numbers. This dramatically shrinks the model's size, reduces memory usage, and speeds up computation, making it cheaper and faster to run on less powerful hardware. While it can sometimes lead to a slight drop in accuracy, techniques like quantization-aware training aim to minimize this impact, sometimes even allowing compressed models to outperform their full-precision originals.