It depends on the underlying process that generates the data.
You can always fit N points exactly with a polynomial of degree N-1. A cubic will fit these points exactly. Often, such a polynomial does not give very good predicitions, though.
The next-best choice is a quadratic, followed by a square root, then a linear model.
_____
In the graph, the model with the highest r²-value is best at predicting the given points. Here are the r² values for the different models.
linear: 0.9595
cubic: 1
square root: 0.9991
quadratic: 0.9992
If the underlying process is likely to produce points that continue to increase at a slower rate, I'd probably choose the square root model as the most likely to be reasonably accurate. If the points are likely to eventually start decreasing, I'd choose a quadratic model.