I haven't checked it myself but you may want to look into popular simulation suites (OrCad, PSD, etc.) to see what they use there. As those models in commercial suites tend to be copyrighted, it is hard to share them in public.
Yes, they are terrible. This thread presents some results showing the simulated ft to be around a factor of 6 low at the nominal bias current. It also shows a PSPICE model which is pretty terrible also. I ended up tweaking the models to get a much better fit to ft vs current. The results are shown later in the thread. Unfortunately, I made the assumption that they couldn't possibly have messed up the DC parameters, but sure enough those are hosed too. This other thread shows some of the problems with the DC parameters. Among these is an RC that's much too high. It causes the internal Vcb to vary with increasing current so much that, combined with the Early effect, it gives a downward tilt to beta vs current.
Unfortunately, I haven't gone back and combined these changes into one model that fixes both the DC and AC parameters. I've been meaning to do it for quite some time now. Maybe this weekend I will bite the bullet and give it a try.
Doing this kind of stuff is no fun at all! The data sheets don't give all the info you need to get all the model parameters. Then there's the matter of eyeballing the data sheet to get tons of data values, then manually entering the data into Excel to plot them. What's really needed are text files containing the data for the characteristic curves, ft vs current, etc. It would be way cool if the data for the Gummel plots could be obtained from the vendor as text files.
Later in that thread, I did some plots of the MJE15030 and MJE15031 models ft vs. current. This post in that thread shows that the plots agree very well with the data sheet. You might give those a try.