In some communities open access journals are the most prestigious journals. In machine learning JMLR (http://jmlr.csail.mit.edu/) is (I think) widely considered the top journal. It was formed when a large group of top machine learning researchers grew fed up with the high cost of paid publication and together resigned from the board of a competing publication (http://jmlr.csail.mit.edu/statement.html). All of the top machine learning conferences (NIPS, ICML, UAI, AIStats, COLT) publish their proceedings online too. I think it's only a matter of time before all other communities follow suit, but some communities may lag behind longer than others. The machine learning community was lucky in that a large group of leading researchers decided to collectively make a clean break and support a new journal--the new journal became the premier journal almost over night so there was no slow transition phase.
