My guess would be us a faster model and less context.