Hi @MoKret, and welcome to the world of cryoEM!
As you may already know, the GSFSC curve measures the correlation between two independent half-sets of particles. This correlation tells us something about how much of what we observe is signal and how much is overfit noise. (I discuss the GSFSC a bit more in this forum post).
However, if there are duplicate copies of particles in each half set, the correlation between the half sets will never go to zero, because these duplicates correlate perfectly with each other! This type of curve, where you get a nice steady drop which flattens out before it reaches zero, is a classic indicator that you have duplicate particles in your particle stack. If you run these particles through a Remove Duplicate Particles job and then reconstruct them, the curve will most likely go to zero as you expect.
I also will draw your attention to our guide page on common plots produced by CryoSPARC, where we discuss this effect and some others you may see as you keep working through your data. You are always welcome to ask questions here, of course!
I hope that helps!