TikZ: difference between \node and \coordinate?

\node requires a caption:

\node (name) at (coordinate) {caption};

\coordinate does not use a caption:

\coordinate (name) at (coordinate);

\node can also have a shape and dimension, \coordinate is just a point

Well, perhaps it's interesting to look at the pgfmanual :

1. \coordinate is a shortcut for \path ... coordinate[⟨options⟩](⟨name⟩)at(⟨coordinate⟩) ...;

2. and it's the same that node[shape=coordinate][]⟨options ⟩](⟨name ⟩)at(⟨coordinate ⟩){}, where the at part might be missing.

   Since nodes are often the only path operation on paths, there are two special commands for creating paths containing only a node: \node Inside {tikzpicture} this is an abbreviation for \path node. \coordinate Inside {tikzpicture} this is an abbreviation for \path coordinate.

3. pgf and TikZ define three shapes, by default:

   • rectangle,
   • circle, and
   • coordinate.

   The coordinate shape is handled in a special way by TikZ. When a node x whose shape is coordinate is used as a coordinate (x), this has the same effect as if you had said (x.center). None of the special “line shortening rules” apply in this case. This can be useful ...

4. finally

   The exact behaviour of shapes differs, shapes defined for more special purposes (like a, say, transistor shape) will have even more custom behaviors. However, there are some options that apply to most shapes.

It's why some default values like inner sep are different.

node introduces an inner sep and hence does not produce a geometrical point in true sense. coordinate or \node [coordinate] on the other hand, does not have inner sep.

\documentclass{article}
\usepackage{tikz}
\begin{document}
\verb|\node|  introduces an \verb|inner sep|:

\begin{tikzpicture}
    \coordinate (cone) at (0,0);
    \node (none) at (4,0) {};
      \draw[->] (cone) -- (none);
    \draw[draw,red,->] (none) -- (0:5);
\end{tikzpicture}

\verb|\node[coordinate]| or \verb|\coordinate|  does not introduce an \verb|inner sep|:

\begin{tikzpicture}
    \coordinate (cone) at (0,0);
    \node[coordinate] (none) at (4,0) {};     %% or \coordinate (none) at (4,0);
      \draw[->] (cone) -- (none);
    \draw[draw,red,->] (none) -- (0:5);
\end{tikzpicture}

If you use \verb|\node|, you may be forced to use \verb|none.center| to get rid of the gap noticed between the two lines:

\begin{tikzpicture}
    \coordinate (cone) at (0,0);
    \node (none) at (4,0) {};
      \draw[->] (cone) -- (none.center);
    \draw[draw,red,->] (none.center) -- (0:5);
\end{tikzpicture}

Hence, when you need a geometric point, use \verb|\node[coordinate]| or \verb|\coordinate|.
\end{document}

Bottom line: When you need a point use coordinate.