Just to clarify, 5.85 ms and 11.8 ms are the data points of the baseline, so complexity is indeed linear there. The Elixir implementation, though, runs in 225.4 ms (39x slower) and 664 ms (56x slower).

Thanks for pointing out the mistake in the data sent to Wolfram Alpha! Still, if you look at the results table, you can see that the factor of increase is ~2x at the beginning, but is ~3x at the end.

(This is why I wanted a plot instead of a table in the blog post :p, because then the curve speaks for itself)