diff --git a/libs/partners/openai/tests/integration_tests/chat_models/test_azure.py b/libs/partners/openai/tests/integration_tests/chat_models/test_azure.py
index 3436f165d7cc1..f5787aea65c1c 100644
--- a/libs/partners/openai/tests/integration_tests/chat_models/test_azure.py
+++ b/libs/partners/openai/tests/integration_tests/chat_models/test_azure.py
@@ -1,5 +1,6 @@
 """Test AzureChatOpenAI wrapper."""
 
+import json
 import os
 from typing import Any, Optional
 
@@ -225,3 +226,51 @@ def test_openai_invoke(llm: AzureChatOpenAI) -> None:
     result = llm.invoke("I'm Pickle Rick", config=dict(tags=["foo"]))
     assert isinstance(result.content, str)
     assert result.response_metadata.get("model_name") is not None
+
+
+def test_json_mode() -> None:
+    llm = AzureChatOpenAI(  # type: ignore[call-arg, call-arg, call-arg]
+        deployment_name="gpt-4o-mini",
+        openai_api_version=OPENAI_API_VERSION,
+        azure_endpoint=OPENAI_API_BASE,
+        openai_api_key=OPENAI_API_KEY,
+    )
+    response = llm.invoke(
+        "Return this as json: {'a': 1}", response_format={"type": "json_object"}
+    )
+    assert isinstance(response.content, str)
+    assert json.loads(response.content) == {"a": 1}
+
+    # Test streaming
+    full: Optional[BaseMessageChunk] = None
+    for chunk in llm.stream(
+        "Return this as json: {'a': 1}", response_format={"type": "json_object"}
+    ):
+        full = chunk if full is None else full + chunk
+    assert isinstance(full, AIMessageChunk)
+    assert isinstance(full.content, str)
+    assert json.loads(full.content) == {"a": 1}
+
+
+async def test_json_mode_async() -> None:
+    llm = AzureChatOpenAI(  # type: ignore[call-arg, call-arg, call-arg]
+        deployment_name="gpt-4o-mini",
+        openai_api_version=OPENAI_API_VERSION,
+        azure_endpoint=OPENAI_API_BASE,
+        openai_api_key=OPENAI_API_KEY,
+    )
+    response = await llm.ainvoke(
+        "Return this as json: {'a': 1}", response_format={"type": "json_object"}
+    )
+    assert isinstance(response.content, str)
+    assert json.loads(response.content) == {"a": 1}
+
+    # Test streaming
+    full: Optional[BaseMessageChunk] = None
+    async for chunk in llm.astream(
+        "Return this as json: {'a': 1}", response_format={"type": "json_object"}
+    ):
+        full = chunk if full is None else full + chunk
+    assert isinstance(full, AIMessageChunk)
+    assert isinstance(full.content, str)
+    assert json.loads(full.content) == {"a": 1}
diff --git a/libs/partners/openai/tests/integration_tests/chat_models/test_base.py b/libs/partners/openai/tests/integration_tests/chat_models/test_base.py
index 3e03755b765bd..f1eb6c39a5f5c 100644
--- a/libs/partners/openai/tests/integration_tests/chat_models/test_base.py
+++ b/libs/partners/openai/tests/integration_tests/chat_models/test_base.py
@@ -1,6 +1,7 @@
 """Test ChatOpenAI chat model."""
 
 import base64
+import json
 from typing import Any, AsyncIterator, List, Literal, Optional, cast
 
 import httpx
@@ -865,3 +866,41 @@ class InvalidJoke(BaseModelProper):
         chat.invoke("Tell me a joke about cats.")
     with pytest.raises(openai.BadRequestError):
         next(chat.stream("Tell me a joke about cats."))
+
+
+def test_json_mode() -> None:
+    llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
+    response = llm.invoke(
+        "Return this as json: {'a': 1}", response_format={"type": "json_object"}
+    )
+    assert isinstance(response.content, str)
+    assert json.loads(response.content) == {"a": 1}
+
+    # Test streaming
+    full: Optional[BaseMessageChunk] = None
+    for chunk in llm.stream(
+        "Return this as json: {'a': 1}", response_format={"type": "json_object"}
+    ):
+        full = chunk if full is None else full + chunk
+    assert isinstance(full, AIMessageChunk)
+    assert isinstance(full.content, str)
+    assert json.loads(full.content) == {"a": 1}
+
+
+async def test_json_mode_async() -> None:
+    llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
+    response = await llm.ainvoke(
+        "Return this as json: {'a': 1}", response_format={"type": "json_object"}
+    )
+    assert isinstance(response.content, str)
+    assert json.loads(response.content) == {"a": 1}
+
+    # Test streaming
+    full: Optional[BaseMessageChunk] = None
+    async for chunk in llm.astream(
+        "Return this as json: {'a': 1}", response_format={"type": "json_object"}
+    ):
+        full = chunk if full is None else full + chunk
+    assert isinstance(full, AIMessageChunk)
+    assert isinstance(full.content, str)
+    assert json.loads(full.content) == {"a": 1}