The network used is a fat b40c384 with a ton of playouts. For the first move it likes tengen and hoshi. After tengen it likes a one space jump or knights move by white. The response by black in the game, the "pendulum" to play close to the white knights move is also the top move.

It's a boring looking game, but it reversed several times in AI analysis in points that don't seem obvious.

First reversal: wB4 hane on the second line is a loss. Not clear to me why, but I guess you keep sente more easily that way and just draw.
Second reversal: bE4 some innocuous exchange by pushing in the middle that lets white fix the shape and make a clamp - not forcing leaves the option of a cut, making moves on the right sente
Third reversal: wH4 coming back around to the side nullified white chances

That last part of the game is the most interesting one, because white can't actually live inside easily, but can use the forcing moves to jump into the right side to win. This is because Black messed up and let white fix the shape. If Black didn't push the middle, black hane on the right would be more threatening and kind of sente

I didn't transcribe the rest of the moves, but at this point white just dies

https://online-go.com/game/25215645

download the sgf and look at it with Lizzie if you want to see big playout numbers

There are so many possible lines on 9x9...
I did a quick overview of main credible lines with KataGo 40b-s5G09 (mainly with 10K to 100K playouts) beginning after a 5-4 move, and you'll see there are mainly lines!
Warning: most a few % winrate from top move but a few ones are less optimal than that, and all interesting moves are not included.

It gets way worse. Lots of lines become a draw even after KataGo rates them 30% win rate -1 point. It just seems black is able to get that last dame after all in the best variation for both.

I had some 3-4 lines where white wins. Later nets were able to tie after opening 3-4 even though it hated it with a passion.