Meh, computers are like that. Don't rely on a computer to tell you what to do for the first couple of moves in a game. They also have a different opening book than you, so they will favor certain openings more.
Computer analysis rated the Caro-Kann's first move as an inaccuracy
Sort:
The Caro-Kann is considered near-bulletproof nowadays. Chess.com's own Sam Shankland and Alex Lenderman employ it against very strong opposition; the latter almost exclusively, being the student of the USA's biggest Caro expert, Giorgi Kacheishvili.
I think every single top player has played it at some point. Anand, Ivanchuk, Topalov and Leko are especially fond of it. But it has also been played by Carlsen, Grischuk, Ponomariov, Svidler, Morozevich, Kramnik, you name it.
Kasparov used the opening a lot in his early days, and even wrote a book on it! Karpov of course used the Caro as his main weapon in the supertournaments of the 1990s.
Don't trust computers too much.
I submitted a game I just played for computer analysis from this site, and the computer rated 1...c6, the fourth most common response to 1.e4, as an inaccuracy. It also suggested 1...e5 as a better alternative. I've played the Caro before, and this has never happened.
Any thoughts? I would imagine that the Caro-Kann is quite sound, and that the computer is glitching.