Re-ran the benchmark from https://gist.github.com/denispeplin/99047e34b0974b00116f335651d551d6 on OTP24

benchmark.ex

# Requires Elixir 1.12+
Mix.install([
  {:benchee, "~> 1.0"}
])

# Benchmark copied from https://gist.github.com/denispeplin/99047e34b0974b00116f335651d551d6
defmodule So.Exclude do
  def exclude1() do
    a = [1, 2, 3, 4]
    b = [3, 4, 5, 6]

    run_benchmarks(a, b)
  end

  def exclude2(size, max_value \\ nil) do
    max_value = max_value || size

    a = random_list(size, max_value)
    b = random_list(size, max_value)

    run_benchmarks(a, b)
  end

  defp run_benchmarks(a, b) do
    Benchee.run(%{
      "enum_filter" => fn ->
        Enum.filter(a, fn el -> !Enum.member?(b, el) end)
      end,
      "substract" => fn ->
        a -- b
      end,
      "mapset_difference" => fn ->
        MapSet.difference(MapSet.new(a), MapSet.new(b)) |> MapSet.to_list
      end,
      "ordsets" => fn ->
        :ordsets.subtract(:ordsets.from_list(a), :ordsets.from_list(b)) |> :ordsets.to_list
      end
    }, time: 10)
  end

  defp random_list(size, max_value) do
    Enum.map(1..size, fn(_size) -> Enum.random(1..max_value) end)
  end
end

results.ex

# Operating System: macOS
# CPU Information: Intel(R) Core(TM) i5-8279U CPU @ 2.40GHz
# Number of Available Cores: 8
# Available memory: 16 GB
# Elixir 1.12.1
# Erlang 24.0.2

iex> So.Exclude.exclude1()

Name                        ips        average  deviation         median         99th %
substract                4.88 M      205.00 ns  ±1386.42%           0 ns        1000 ns
ordsets                  3.03 M      330.04 ns ±10483.97%           0 ns        1000 ns
enum_filter              2.93 M      340.95 ns ±18768.28%           0 ns        1000 ns
mapset_difference        1.38 M      725.01 ns  ±7031.83%        1000 ns        1000 ns

Comparison: 
substract                4.88 M
ordsets                  3.03 M - 1.61x slower +125.04 ns
enum_filter              2.93 M - 1.66x slower +135.95 ns
mapset_difference        1.38 M - 3.54x slower +520.02 ns

iex> So.Exclude.exclude2(100)

Name                        ips        average  deviation         median         99th %
substract              229.89 K        4.35 μs   ±433.75%           4 μs          20 μs
ordsets                215.92 K        4.63 μs   ±628.66%           4 μs          22 μs
enum_filter             97.18 K       10.29 μs   ±213.13%           9 μs          32 μs
mapset_difference       47.86 K       20.89 μs    ±56.83%          19 μs          64 μs

Comparison: 
substract              229.89 K
ordsets                215.92 K - 1.06x slower +0.28 μs
enum_filter             97.18 K - 2.37x slower +5.94 μs
mapset_difference       47.86 K - 4.80x slower +16.54 μs

iex> So.Exclude.exclude2(1_000)

Name                        ips        average  deviation         median         99th %
substract                7.32 K      136.63 μs    ±22.26%         124 μs         254 μs
ordsets                  6.66 K      150.10 μs    ±32.38%         133 μs         308 μs
mapset_difference        2.75 K      364.27 μs    ±19.44%         335 μs         620 μs
enum_filter              0.98 K     1016.90 μs    ±31.22%         842 μs     1871.60 μs

Comparison: 
substract                7.32 K
ordsets                  6.66 K - 1.10x slower +13.48 μs
mapset_difference        2.75 K - 2.67x slower +227.64 μs
enum_filter              0.98 K - 7.44x slower +880.28 μs

iex> So.Exclude.exclude2(100_000)

Name                        ips        average  deviation         median         99th %
ordsets                   24.55       40.74 ms    ±24.12%       37.26 ms       75.39 ms
substract                 24.10       41.50 ms     ±4.97%       41.50 ms       46.66 ms
mapset_difference         17.36       57.60 ms     ±4.80%       57.72 ms       65.84 ms
enum_filter              0.0775    12909.30 ms     ±0.00%    12909.30 ms    12909.30 ms

Comparison: 
ordsets                   24.55
substract                 24.10 - 1.02x slower +0.76 ms
mapset_difference         17.36 - 1.41x slower +16.86 ms
enum_filter              0.0775 - 316.89x slower +12868.56 ms