Playing on the side as early as move #5 is not the AI gospel. KataGo evaluates it as a 2% / 0.7 point loss. At the time of this review, this is considered to be in the margin of error of AI analysis, so we can't call the san ren sei bad.
It should come as no surprise that KataGo recommends , the early 3-3 invasion. Next is a way to take sente and retaliate with . The particularity of the san ren sei doesn't have an apparent influence on KataGo's fuseki. It's early 3-3 all over again.
With LeelaZero it's a little different: she plays at a
When suggesting the approach of to KataGo, it brings about this sequence. White plays a conventional looking stabilizing move at a next.
This loses a marginal amount of points or percentage points with respect to the early 3-3.
LeelaZero ignores to 3-3 the diagonally opposite corner. White will take sente to come back here and 3-3 at , now with support.
If pincers, White is happy (according to LeelaZero) to jump into the corner. is notable. After , Black will play a, which is may adhere to the traditional framework building objective of the san ren sei.
KataGo likes this variation a little better. After (or even after ), Black takes sente for invading at a. The idea of building a framework is abandoned.