my tests are against the Apple Intelligence model, fwiw, not any earlier model