diff --git a/ai_eval/llm_services.py b/ai_eval/llm_services.py
index 84dcdee..59734d4 100644
--- a/ai_eval/llm_services.py
+++ b/ai_eval/llm_services.py
@@ -136,6 +136,45 @@ def _get_headers(self):
         self._ensure_token()
         return {'Authorization': f'Bearer {self._access_token}'}
 
+    @staticmethod
+    def _parse_models_field(raw_models):
+        """
+        Parse the value of a top-level "models" field into a list of model ids/names.
+        """
+        if isinstance(raw_models, list):
+            return [str(m) for m in raw_models]
+        if isinstance(raw_models, str):
+            return [str(raw_models)]
+        if isinstance(raw_models, dict):
+            parsed_models = []
+            for key, val in raw_models.items():
+                if isinstance(val, dict):
+                    candidate = val.get("name") or val.get("id") or key
+                elif isinstance(val, str) and val.strip():
+                    candidate = val
+                else:
+                    candidate = key
+                parsed_models.append(str(candidate))
+            return parsed_models
+        return []
+
+    @classmethod
+    def _parse_models_response(cls, data):
+        """
+        Parse a models endpoint JSON response into a list of model ids/names.
+        """
+        if isinstance(data, dict):
+            if "models" in data:
+                return cls._parse_models_field(data["models"])
+            if isinstance(data.get("data"), list):
+                return [str(m.get("id", str(m))) for m in data["data"]]
+            return []
+        if isinstance(data, list):
+            return [str(m) for m in data]
+        if isinstance(data, str):
+            return [str(data)]
+        return []
+
     def get_response(
             self,
             model,
@@ -196,20 +235,7 @@ def get_available_models(self):
             response.raise_for_status()
             data = response.json()
 
-            models = []
-
-            if isinstance(data, dict):
-                if "models" in data:
-                    if isinstance(data["models"], list):
-                        models = [str(m) for m in data["models"]]
-                    elif isinstance(data["models"], str):
-                        models = [str(data["models"])]
-                elif "data" in data and isinstance(data["data"], list):
-                    models = [str(m.get("id", str(m))) for m in data["data"]]
-            elif isinstance(data, list):
-                models = [str(m) for m in data]
-            elif isinstance(data, str):
-                models = [str(data)]
+            models = self._parse_models_response(data)
 
             # Filter out non-string model names and empty strings
             models = [m for m in models if isinstance(m, str) and m.strip()]
diff --git a/ai_eval/tests/test_ai_eval.py b/ai_eval/tests/test_ai_eval.py
index 313bcfa..e5666b8 100644
--- a/ai_eval/tests/test_ai_eval.py
+++ b/ai_eval/tests/test_ai_eval.py
@@ -19,6 +19,7 @@
 )
 from ai_eval.base import AIEvalXBlock
 from ai_eval.supported_models import SupportedModels
+from ai_eval.llm_services import CustomLLMService
 from ai_eval.backends.factory import BackendFactory
 from ai_eval.backends.judge0 import Judge0Backend
 from ai_eval.backends.custom import CustomServiceBackend
@@ -220,6 +221,33 @@ def test_multiagent_block_evaluator_response():
     assert resp["is_evaluator"]
 
 
+def test_custom_llm_models_dict_response_parsed():
+    """Custom service supports {"models": {id: {...}}} response bodies."""
+    service = CustomLLMService(
+        models_url="https://example.com/models",
+        completions_url="https://example.com/completions",
+        token_url="https://example.com/token",
+        client_id="client",
+        client_secret="secret",
+    )
+    service._get_headers = Mock(return_value={"Authorization": "Bearer token"})  # pylint: disable=protected-access
+
+    mocked_response = Mock()
+    mocked_response.json.return_value = {
+        "models": {
+            "Meta-Llama4-Maverick": {
+                "name": "Meta-Llama4-Maverick",
+                "display_name": "Maverick (Llama 4)",
+            },
+            "gpt-4o": {"display_name": "GPT-4o"},
+        }
+    }
+    mocked_response.raise_for_status.return_value = None
+
+    with patch("ai_eval.llm_services.requests.get", return_value=mocked_response):
+        assert service.get_available_models() == ["Meta-Llama4-Maverick", "gpt-4o"]
+
+
 @pytest.mark.parametrize(
     "backend_config, expected_backend_class",
     [