diff --git a/bench/bench.py b/bench/bench.py
new file mode 100644
index 0000000..ac726c2
--- /dev/null
+++ b/bench/bench.py
@@ -0,0 +1,124 @@
+from cryptomite.dodis import Dodis
+from cryptomite.toeplitz import Toeplitz
+from cryptomite.utils import von_neumann
+from cryptomite.trevisan import Trevisan 
+from numpy.random import randint
+import numpy as np
+
+from time import time
+
+import pandas as pd
+
+
+cache1 = dict()
+cache2 = dict()
+
+N_REPEATS = 5
+
+seed_caches = [dict() for _ in range(N_REPEATS)]
+cache3 = dict()
+def get_random(cache, n):
+    if n not in cache:
+        cache[n] = randint(0, 2, size=n)
+
+    return cache[n]
+
+# compare with naive
+
+inp_lens = [x * 10 ** n for x in [10, 18, 32, 56] for n in range(1, 10)]
+# inp_lens = [10 ** 8]
+names = ["Extractor", "Input Length", "Ratio"]
+#ratios = (0.5, 0.75)
+ratios = [0.5]
+columns = ["Wait (s)"]
+indices = []
+rows = []
+
+# Dodis
+print("\nDodis: ", end="", flush=True)
+for inp_len in inp_lens:
+    inp1 = get_random(cache1, inp_len)
+    inp2 = get_random(cache2, inp_len)
+
+    for ratio in ratios:
+        waits = []
+        for i in range(N_REPEATS):
+            inp2 = get_random(seed_caches[i], inp_len)
+            out_len = int(inp_len * ratio)
+            ext = Dodis(inp_len, out_len)
+            start = time()
+            ext.extract(inp1, inp2)
+            wait = time() - start
+            waits.append(wait)
+            print(".", end="", flush=True)
+        print("|", end="", flush=True)
+        rows.append(sum(waits) / len(waits))
+        indices.append(('Dodis', inp_len, ratio))
+
+# VN
+print("\nVN: ", end="", flush=True)
+for inp_len in inp_lens:
+    waits = []
+    for i in range(N_REPEATS):
+        inp1 = get_random(seed_caches[i], inp_len)
+        start = time()
+        von_neumann(inp1)
+        wait = time() - start
+        waits.append(wait)
+        print(".", end="", flush=True)
+    print("|", end="", flush=True)
+    rows.append(sum(waits) / len(waits))
+    indices.append(('Von Neumann', inp_len, np.nan))
+
+# Toeplitz
+print("\nToeplitz", end="", flush=True)
+for inp_len in inp_lens:
+    inp1 = get_random(cache1, inp_len)
+
+    for ratio in ratios:
+        out_len = int(inp_len * ratio)
+        waits = []
+        for i in range(N_REPEATS):
+            seed = get_random(seed_caches[i], inp_len + out_len - 1)
+            ext = Toeplitz(inp_len, out_len)
+            start = time()
+            ext.extract(inp1, seed)
+            wait = time() - start
+            waits.append(wait)
+            print(".", end="", flush=True)
+        print("|", end="", flush=True)
+        rows.append(sum(waits) / len(waits))
+        indices.append(('Toeplitz', inp_len, ratio))
+
+# Trevisan
+print("\nTrevisan: ", end="", flush=True)
+for inp_len in inp_lens:
+    if inp_len > 10000:
+        continue
+    inp1 = get_random(cache1, inp_len)
+
+    for ratio in ratios:
+        min_ent = int(inp_len * ratio)
+        ext = Trevisan(inp_len, min_ent, 2 ** -20)
+        seed_length = ext.ext.get_seed_length()
+        waits = []
+        for i in range(N_REPEATS):
+            seed = get_random(seed_caches[i], seed_length)
+            start = time()
+            ext.extract(inp1, seed)
+            wait = time() - start
+            waits.append(wait)
+            print(".", end="", flush=True)
+        rows.append(sum(waits) / len(waits))
+        indices.append(('Trevisan', inp_len, ratio))
+print("")
+
+index = pd.MultiIndex.from_tuples(indices, names=names)
+
+df = pd.DataFrame(rows, index=index, columns=columns)
+
+print(df)
+
+ndf = df.copy().reset_index()
+ndf.loc[ndf["Extractor"] == "Von Neumann", "Ratio"] = 0.5
+ndf["Rate"] = ndf["Input Length"] * ndf["Ratio"] / ndf["Wait (s)"]