Looking at some old Sabato articles, the biggest one they
got wrong on election day in 2016 was calling WI Likely D (whereas PA and MI were Lean D). They also labeled a bunch of Trump states as Lean D, though it's worth noting that they didn't use the Tossup or Tilt categories.
They
got all but two states right in 2012: they wrongly predicted FL and VA(!) would go for Romney, having both as Lean R (and they specifically noted that VA was really a complete tossup, alongside Colorado and New Hampshire).
They
got all but two things right in 2008: they wrongly said that McCain would win NE-2 and Indiana. But credit to them for this statement: "If Barack Obama wins the popular vote by 7 or 8 percentage points, then he will also carry at least a couple of the following states: Arizona, Georgia, Indiana, Montana, North Dakota."
In 2004, they
got two states wrong: they called Florida for Democrats, and Wisconsin for Republicans, but in both cases they stated that their confidence was "low".
So for Sabato for these elections, it seems like the two biggest surprises were WI in 2016 and IN in 2008.