Mind the Gap! Static and Interactive Evaluations of Large Audio Models