The University of Michigan Predictor (http://vmr.engin.umich.edu/Model/_balloon/index.py) is currently favored to United States users. This is because it uses the Rapid Refresh (RR) model which is only for the US. We choose not to use GFS because it has a larger refresh time and has an altitude ceiling of ~60,000 feet whereas the RR gives us new data each hour and up to ~95,000 feet.
Our ascent and descent calculations are based on user inputs. So instead of using a separate calculator to figure this out, we ask you for your balloon weight, helium intake (based on our standard tank size), parachute diameter, and payload weight.
Once you input your launch location, we sniff out new weather stations along the path instead of using only the weather station closest to launch to increase the accuracy. With each packet that is received, we can update the predicted path in real time so you can chase the landing and not the balloon. This is also great for terminating a flight if the predicted model was wayyy off. We've had to do this. Darn models! That said, our last flight had a landing error of only about 1.8 miles, so we are very happy with accuracy of the predictor.
We will be implementing the option of GFS, NAM, EURO, RR, etc. so you can use this globally and compare and contrast models - similar to how hurricane paths are predicted. The main difference is that some of these models have very low ceilings compared to how high a HAB can actually reach. While one may argue that 60,000 feet is the edge of the jet stream with relatively calm winds, that varies day to day, season to season, and can make the difference between landing on a farm and landing in a forest/lake/city.
We have a student team working with our professor to make the predictor both easier to use, and more accurate, so we'd be happy to hear about other model constraints so we can continue to improve on this!