The model currently puts the president at just a 1-in-20 chance to win the Electoral College, but you shouldn’t believe it. Don’t take my word for it; read this post from Andrew Gelman, a statistician who helped create the model and doesn’t believe it himself.
There’s a lot of technical detail in the full post, but here’s the key bit for the casual reader:
Our backfitted predictions seem to be overconfident . . . so I don’t fully believe our numbers either.
. . . Our model was not designed to make sense in the tails; we didn’t think through what would happen if the vote share needed for Trump to win became a tail event, or consider the fact that this was inevitable if Biden were to stay steady in the polls.
So, when the model puts a chance as low as 5 percent or so, it’s not reliable. Oops.