My impression is that Leibnitz usually offers several approaches that are alternatives, and are not consistent. (Not surprising, since most of the papers are his notes, written "for the desk", not intended by him for publication.)
Wallis in his book on integration, which is prior both Newton and Leibnitz, uses the concept of a right hand limit of a quantity that goes to zero. So it's arbitrarily small but never null.
Leibnitz in a published short paper, on what infinitesimals are, constructs them in a similar (but not always quite the same) way. Perhaps this may be taken as his "official" opinion, seeing that he published it. (But he treats them inconsistently across his entire body of work.)
Basically, he constructs a right triangle in cartesian coordinates, and intersects it with a geodesic in two places. He moves the geodesic continuously toward a vertex of the triangle, so creating another triangle whose sides get shorter and converge to zero (if the geodesic touches the vertex). But he defines a fraction where the numbers are lengths of the sides of the triangle so constructed.
In this case: the denominator cannot be zero, if the system is to be consistent, so the motion of the geodesic is restricted in that it cannot touch the vertex. Whatever can be constructed by such a motion is an infinitesimal, he says.
He defines each infinitesimal as a fraction constructed by such motion. Both denominator and numerator become smaller as the geodesic approaches, but never become null. And because the angle of the geodesic where it intersects each side of the triangle is NOT in general the same, the sides of the triangle constructed are NOT, except in a special case of the construction, equal.
Obviously, not all infinitesimals are identical. Infinitesimal to him, therefore, as far as his published work is concerned, refers to any quantity always decreasing, in the same way as Hinchin in his 1950's calculus textbook presented the matter. To be an infinitesimal, as far as Leibnitz is concerned, is to be constructed in a certain way by a certain series of changes in another function, but it says nothing of the quantity itself. (Hinchin nods in approval.)
However, in his discussion of modelling quality by quantity, via continuity, infinitesimals are arbitrarily small quantities, eventually becoming zero, at which point, one quality transforms into another.
I suggest treating his opinion of the ontology of infinitesimals as not substantially different from that of Wallis (limit of a variable quantity, going to zero, from the right hand side) which presumably inspired it, and is based on right hand limits of functions that decrease as their input increases. Merely he is not consistent across all his papers.
EDIT:
I am looking through one of my archive trying to find my translation of the important passages of Wallis's Arithmetica Infinitorum. From memory, Wallis wrote in one place the equivalent of
$\underset{x\rightarrow 0+}{\lim}x$
as what he meant by infinitesimal, in modern notation. This is the thickness of lines in his diagrams. Summing an infinity of such lines required to fill the figures gave the areas of the figures. But, according to Beeley, in his later communications to Leibniz, he writes they -- the lines -- have literally null thickness, he has decided. Leibniz disagrees. Leibniz's own idea in modern notation is closer to Wallis's first statement, and more precisely the above conception. Where the $x$ is input to a function (the infinitesimal), such that $dx = \frac{f(x)}{g(x)}\rightarrow 0+$ there where $x\rightarrow 0+$.
