A larger window is also not the same as better recall inside it. Models can attend unevenly across a very long input, and a detail buried at the start of a hundred thousand tokens may be treated less reliably than the same detail near the end. When a task really depends on one paragraph, saying which paragraph is cheaper and more reliable than hoping the whole file fits.
Watch the difference between "advertised" and "usable." A vendor may quote a window that applies only to certain endpoints, or a plugin may quietly truncate input before it reaches the model. See how to read a model listing.