Can small AI models think as well as large ones?