diff --git a/src/lighteval/metrics/utils/llm_as_judge.py b/src/lighteval/metrics/utils/llm_as_judge.py
index 989588029..a12a3e4fd 100644
--- a/src/lighteval/metrics/utils/llm_as_judge.py
+++ b/src/lighteval/metrics/utils/llm_as_judge.py
@@ -344,7 +344,7 @@ def __call_api(prompt):
                         "response_format": self.response_format,
                     }
                     if max_new_tokens is not None:
-                        kwargs["max_tokens"] = (max_new_tokens,)
+                        kwargs["max_tokens"] = max_new_tokens
                     if self.api_key is not None:
                         kwargs["api_key"] = self.api_key.get_secret_value()
                     if self.url is not None:
diff --git a/tests/unit/metrics/test_llm_as_judge_litellm.py b/tests/unit/metrics/test_llm_as_judge_litellm.py
new file mode 100644
index 000000000..f3a4b75f1
--- /dev/null
+++ b/tests/unit/metrics/test_llm_as_judge_litellm.py
@@ -0,0 +1,48 @@
+# MIT License
+#
+# Copyright (c) 2024 The HuggingFace Team
+
+from types import SimpleNamespace
+
+from lighteval.metrics.utils.llm_as_judge import JudgeLM
+
+
+class _FakeLiteLLM:
+    drop_params = False
+    cache = None
+
+    def __init__(self):
+        self.max_tokens = None
+
+    @staticmethod
+    def supports_reasoning(model):
+        return False
+
+    def completion(self, **kwargs):
+        self.max_tokens = kwargs.get("max_tokens")
+        message = SimpleNamespace(content="1")
+        choice = SimpleNamespace(message=message)
+        return SimpleNamespace(choices=[choice])
+
+
+def test_litellm_judge_sends_max_tokens_as_int(monkeypatch):
+    fake = _FakeLiteLLM()
+    monkeypatch.setitem(__import__("sys").modules, "litellm", fake)
+
+    judge = JudgeLM(
+        model="openai/test-model",
+        templates=lambda question, answer, options=None, gold=None, **kw: [
+            {"role": "user", "content": question}
+        ],
+        process_judge_response=lambda response: response,
+        judge_backend="litellm",
+        max_tokens=64,
+        backend_options={"caching": False, "increase_max_tokens_for_reasoning": False},
+    )
+
+    scores, _, responses = judge.evaluate_answer_batch(["q"], ["a"], [None], [None])
+
+    assert fake.max_tokens == 64
+    assert isinstance(fake.max_tokens, int)
+    assert scores == ["1"]
+    assert responses == ["1"]