A smaller, more efficient model designed to run quickly and use less memory than larger alternatives, often with some trade-off in reasoning capability.