diff --git a/src/strands/models/litellm.py b/src/strands/models/litellm.py
index 005eed3df..9b16c9ff7 100644
--- a/src/strands/models/litellm.py
+++ b/src/strands/models/litellm.py
@@ -205,12 +205,21 @@ async def structured_output(
         Yields:
             Model events with the last being the structured output.
         """
-        if not supports_response_schema(self.get_config()["model_id"]):
+        # Allow LiteLLM proxy usage to bypass supports_response_schema check.
+        # If a proxy is in use, we cannot reliably detect support by model_id alone,
+        # so prefer to let litellm/proxy handle response_format negotiation.
+        client_args = getattr(self, "client_args", {}) or {}
+        use_proxy = bool(client_args.get("use_litellm_proxy") or client_args.get("use_proxy"))
+        model_id = self.get_config().get("model_id")
+
+        if not use_proxy and (not model_id or not supports_response_schema(model_id)):
             raise ValueError("Model does not support response_format")
 
+        logger.debug("calling litellm.acompletion model=%s use_proxy=%s", model_id, use_proxy)
+
         response = await litellm.acompletion(
             **self.client_args,
-            model=self.get_config()["model_id"],
+            model=model_id,
             messages=self.format_request(prompt, system_prompt=system_prompt)["messages"],
             response_format=output_model,
         )
diff --git a/tests/strands/models/test_litellm.py b/tests/strands/models/test_litellm.py
index bc81fc819..e1b4cf15f 100644
--- a/tests/strands/models/test_litellm.py
+++ b/tests/strands/models/test_litellm.py
@@ -332,3 +332,33 @@ def test_tool_choice_none_no_warning(model, messages, captured_warnings):
     model.format_request(messages, tool_choice=None)
 
     assert len(captured_warnings) == 0
+
+
+class TestModel(pydantic.BaseModel):
+    name: str
+
+
+@pytest.mark.asyncio
+async def test_structured_output_with_proxy_bypasses_support_check(monkeypatch):
+    # Model configured with proxy, supports_response_schema returns False
+    model = LiteLLMModel(client_args={"api_key": "X", "use_litellm_proxy": True}, model_id="some/model")
+
+    # Force supports_response_schema to False to simulate "unknown" capability
+    monkeypatch.setattr(strands.models.litellm, "supports_response_schema", lambda _mid: False)
+
+    # Mock litellm.acompletion to return a response with a choice finishing with tool_calls
+    mock_choice = unittest.mock.Mock()
+    mock_choice.finish_reason = "tool_calls"
+    mock_choice.message = unittest.mock.Mock()
+    mock_choice.message.content = '{"name":"proxy-result"}'
+    mock_response = unittest.mock.Mock()
+    mock_response.choices = [mock_choice]
+
+    async def fake_acompletion(**kwargs):
+        return mock_response
+
+    monkeypatch.setattr(strands.models.litellm.litellm, "acompletion", fake_acompletion)
+
+    stream = model.structured_output(TestModel, [{"role": "user", "content": [{"text": "x"}]}])
+    events = [e async for e in stream]
+    assert events[-1] == {"output": TestModel(name="proxy-result")}