diff --git a/python/requirements.txt b/python/requirements.txt
index 086b203a6d..d1a267f64f 100644
--- a/python/requirements.txt
+++ b/python/requirements.txt
@@ -3,4 +3,3 @@ future
 decorator
 opm
 numpy
-tensorflow-cpu
diff --git a/python/test/ml/hybrid_newton/test_hybrid_newton.py b/python/test/ml/hybrid_newton/test_hybrid_newton.py
index 839e5b501c..e032683f05 100644
--- a/python/test/ml/hybrid_newton/test_hybrid_newton.py
+++ b/python/test/ml/hybrid_newton/test_hybrid_newton.py
@@ -4,10 +4,10 @@ from pathlib import Path
 import numpy as np
 import subprocess
 
+from .utils import Dense, Sequential
 from .utils import collect_input_features, compute_output_vars, write_config, extract_newtit_from_file, extract_unrst_variables
 from .utils import (ABSOLUTE_CASES, RELATIVE_CASES, FEATURE_ENGINEERING_CASES, SCALING_CASES, MULTI_MODEL_CASES, ZERO_NEWTON_CASES, ALL_CASES)
 
-import tensorflow as tf
 from kerasify import export_model
 
 
@@ -89,9 +89,8 @@ class TestHybridNewton(unittest.TestCase):
             input_flat = input_flat.reshape(-1)
             output_flat = output_flat.reshape(-1)
 
-            model = tf.keras.Sequential([
-                tf.keras.Input(shape=(input_flat.shape[0],)),
-                tf.keras.layers.Dense(output_flat.shape[0], use_bias=True)
+            model = Sequential([
+                Dense(input_dim=input_flat.shape[0], output_dim=output_flat.shape[0])
             ])
             model.layers[0].set_weights([np.zeros((input_flat.shape[0], output_flat.shape[0])), output_flat])
             apply_time = self.times[start_idx]
@@ -197,3 +196,5 @@ class TestHybridNewton(unittest.TestCase):
                 f"got {h}, expected <= {b}"
             )
 
+if __name__ == "__main__":
+    unittest.main()
diff --git a/python/test/ml/hybrid_newton/utils.py b/python/test/ml/hybrid_newton/utils.py
index 1ba59b1971..20d2468a30 100644
--- a/python/test/ml/hybrid_newton/utils.py
+++ b/python/test/ml/hybrid_newton/utils.py
@@ -5,6 +5,27 @@ from pathlib import Path
 
 from opm.io.ecl import ERst, ESmry
 
+# Avoid tensorflow dependency but maintain compability with kerasify export_model method
+class Dense:
+    def __init__(self, input_dim, output_dim, activation='linear'):
+        self.weights = np.zeros((input_dim, output_dim), dtype=np.float32)
+        self.biases = np.zeros((output_dim,), dtype=np.float32)
+        self.activation = activation
+
+    def set_weights(self, w_b):
+        self.weights = w_b[0].astype(np.float32)
+        self.biases = w_b[1].astype(np.float32)
+
+    def get_weights(self):
+        return [self.weights, self.biases]
+
+    def get_config(self):
+        return {'activation': self.activation}
+
+class Sequential:
+    def __init__(self, layers):
+        self.layers = layers
+
 
 def extract_unrst_variables(data_dir, deck_file):
     """