diff --git a/bindings/python/src/models.rs b/bindings/python/src/models.rs
index b53a7731..c35ae5a9 100644
--- a/bindings/python/src/models.rs
+++ b/bindings/python/src/models.rs
@@ -700,56 +700,59 @@ impl PyWordPiece {
 #[text_signature = "(self, vocab, unk_token)"]
 pub struct PyWordLevel {}
 
-impl PyWordLevel {
-    fn get_unk(kwargs: Option<&PyDict>) -> PyResult<String> {
-        let mut unk_token = String::from("<unk>");
-
-        if let Some(kwargs) = kwargs {
-            for (key, val) in kwargs {
-                let key: &str = key.extract()?;
-                match key {
-                    "unk_token" => unk_token = val.extract()?,
-                    _ => println!("Ignored unknown kwargs option {}", key),
-                }
-            }
-        }
-        Ok(unk_token)
-    }
-}
-
 #[pymethods]
 impl PyWordLevel {
+    #[getter]
+    fn get_unk_token(self_: PyRef<Self>) -> String {
+        let super_ = self_.as_ref();
+        let model = super_.model.read().unwrap();
+        if let ModelWrapper::WordLevel(ref wl) = *model {
+            wl.unk_token.clone()
+        } else {
+            unreachable!()
+        }
+    }
+
+    #[setter]
+    fn set_unk_token(self_: PyRef<Self>, unk_token: String) {
+        let super_ = self_.as_ref();
+        let mut model = super_.model.write().unwrap();
+        if let ModelWrapper::WordLevel(ref mut wl) = *model {
+            wl.unk_token = unk_token;
+        }
+    }
+
     #[new]
-    #[args(kwargs = "**")]
-    fn new(vocab: Option<PyVocab>, kwargs: Option<&PyDict>) -> PyResult<(Self, PyModel)> {
-        let unk_token = PyWordLevel::get_unk(kwargs)?;
+    #[args(unk_token = "None")]
+    fn new(vocab: Option<PyVocab>, unk_token: Option<String>) -> PyResult<(Self, PyModel)> {
+        let mut builder = WordLevel::builder();
 
         if let Some(vocab) = vocab {
-            let model = match vocab {
-                PyVocab::Vocab(vocab) => WordLevel::builder()
-                    .vocab(vocab)
-                    .unk_token(unk_token)
-                    .build()
-                    .expect("Can only fail when loading from files"),
+            match vocab {
+                PyVocab::Vocab(vocab) => {
+                    builder = builder.vocab(vocab);
+                }
                 PyVocab::Filename(vocab_filename) => {
                     deprecation_warning(
                         "0.9.0",
                         "WordLevel.__init__ will not create from files anymore, \
                             try `WordLevel.from_file` instead",
                     )?;
-                    WordLevel::from_file(vocab_filename, unk_token).map_err(|e| {
-                        exceptions::PyException::new_err(format!(
-                            "Error while loading WordLevel: {}",
-                            e
-                        ))
-                    })?
+                    builder = builder.files(vocab_filename.to_string());
                 }
             };
-
-            Ok((PyWordLevel {}, model.into()))
-        } else {
-            Ok((PyWordLevel {}, WordLevel::default().into()))
         }
+        if let Some(unk_token) = unk_token {
+            builder = builder.unk_token(unk_token);
+        }
+
+        Ok((
+            PyWordLevel {},
+            builder
+                .build()
+                .map_err(|e| exceptions::PyException::new_err(e.to_string()))?
+                .into(),
+        ))
     }
 
     /// Read a :obj:`vocab.json`
@@ -790,17 +793,20 @@ impl PyWordLevel {
     /// Returns:
     ///     :class:`~tokenizers.models.WordLevel`: And instance of WordLevel loaded from file
     #[classmethod]
-    #[args(kwargs = "**")]
+    #[args(unk_token = "None")]
     fn from_file(
         _cls: &PyType,
         py: Python,
         vocab: &str,
-        kwargs: Option<&PyDict>,
+        unk_token: Option<String>,
     ) -> PyResult<Py<Self>> {
         let vocab = WordLevel::read_file(vocab).map_err(|e| {
             exceptions::PyException::new_err(format!("Error while reading WordLevel file: {}", e))
         })?;
-        Py::new(py, PyWordLevel::new(Some(PyVocab::Vocab(vocab)), kwargs)?)
+        Py::new(
+            py,
+            PyWordLevel::new(Some(PyVocab::Vocab(vocab)), unk_token)?,
+        )
     }
 }
 
diff --git a/bindings/python/tests/bindings/test_models.py b/bindings/python/tests/bindings/test_models.py
index 3a118f60..3cdf725d 100644
--- a/bindings/python/tests/bindings/test_models.py
+++ b/bindings/python/tests/bindings/test_models.py
@@ -120,3 +120,12 @@ class TestWordLevel:
             assert isinstance(WordLevel(roberta_files["vocab"]), Model)
         with pytest.deprecated_call():
             assert isinstance(WordLevel(roberta_files["vocab"]), WordLevel)
+
+    def test_can_modify(self):
+        model = WordLevel(unk_token="<oov>")
+
+        assert model.unk_token == "<oov>"
+
+        # Modify these
+        model.unk_token = "<unk>"
+        assert model.unk_token == "<unk>"
diff --git a/tokenizers/src/models/wordlevel/mod.rs b/tokenizers/src/models/wordlevel/mod.rs
index 5d1a4a1a..1ac2a119 100644
--- a/tokenizers/src/models/wordlevel/mod.rs
+++ b/tokenizers/src/models/wordlevel/mod.rs
@@ -107,7 +107,7 @@ impl WordLevelBuilder {
 pub struct WordLevel {
     vocab: HashMap<String, u32>,
     vocab_r: HashMap<u32, String>,
-    unk_token: String,
+    pub unk_token: String,
 }
 
 impl std::fmt::Debug for WordLevel {