Users reported speed improvements of on large models compared to previous iterations. This is particularly noticeable in:
