There has been a lot of discussion recently on DIYDrones about GPS accuracy, the reliability of autonomous flight modes like Return to Home, and safe flying conditions. These are all valuable discussions to be had, but it is also is helpful to consider how well GPS-based positioning works in the best of conditions.
To generate the plot above, I launched and RTLed Solo 26 times and recorded the landing position with tape. These flights were conducted on a flat rooftop in Berkeley with a totally unobstructed view of the Southern sky. I typically had 8-12 satellites for each one of these flights.
I then took an image from the air using Solo with a flat-lensed GoPro and measured the angle and distance from the takeoff zone in ImageJ.
Over these 26 landings, the maximum deviation was over 6 feet and the mean was around 3 feet. I hope this is helpful for those new pilots in the community who are just beginning to experiment.