diff --git a/examples/10_Agentic_Inference/README.md b/examples/10_Agentic_Inference/README.md
index ab3673b51..e8eb565d9 100644
--- a/examples/10_Agentic_Inference/README.md
+++ b/examples/10_Agentic_Inference/README.md
@@ -19,7 +19,8 @@ contiguous and ordered by increasing `turn`.
 
 Required fields are `conversation_id`, `turn`, and `role`. User rows normally
 include `content`; agentic rows can also include `system`, `tools`,
-`tool_calls`, `tool_results`, `reasoning_content`, and `delay_seconds`.
+`tool_calls`, `tool_results`, `reasoning_content`, `reasoning`, and
+`delay_seconds`.
 
 Place the dataset under `examples/10_Agentic_Inference/datasets/` or point the YAML at
 another accessible JSONL path.
diff --git a/src/inference_endpoint/dataset_manager/agentic_inference_dataset.py b/src/inference_endpoint/dataset_manager/agentic_inference_dataset.py
index aa054d15d..dbae868f8 100644
--- a/src/inference_endpoint/dataset_manager/agentic_inference_dataset.py
+++ b/src/inference_endpoint/dataset_manager/agentic_inference_dataset.py
@@ -147,6 +147,20 @@ def _build_conversation_metadata(
 
     for _, row in sorted_group.iterrows():
         role = row.get("role")
+        reasoning_content = row.get("reasoning_content")
+        reasoning = row.get("reasoning")
+        if (
+            reasoning_content is not None
+            and not (
+                isinstance(reasoning_content, float) and pd.isna(reasoning_content)
+            )
+            and reasoning is not None
+            and not (isinstance(reasoning, float) and pd.isna(reasoning))
+        ):
+            raise InputValidationError(
+                f"conversation {row.get('conversation_id')!r} turn {row.get('turn')} "
+                "has both 'reasoning_content' and 'reasoning' values"
+            )
 
         # Format this row into message(s) using the same field extraction as before.
         expanded = _expand_tool_results(row)
@@ -161,6 +175,7 @@ def _build_conversation_metadata(
                 "tool_calls",
                 "tool_results",
                 "reasoning_content",
+                "reasoning",
             ):
                 val = row.get(key)
                 if val is not None and not (isinstance(val, float) and pd.isna(val)):
diff --git a/src/inference_endpoint/openai/openai_msgspec_adapter.py b/src/inference_endpoint/openai/openai_msgspec_adapter.py
index 63318c88e..d639cd949 100644
--- a/src/inference_endpoint/openai/openai_msgspec_adapter.py
+++ b/src/inference_endpoint/openai/openai_msgspec_adapter.py
@@ -49,13 +49,15 @@
 
 def _chat_message_from_dict(msg: dict) -> "ChatMessage":
     """Build a ChatMessage from a dict, forwarding all supported fields."""
+    reasoning_value = msg.get("reasoning_content") or msg.get("reasoning")
     return ChatMessage(
         role=msg["role"],
         content=msg.get("content"),
         name=msg.get("name"),
         tool_calls=msg.get("tool_calls"),
         tool_call_id=msg.get("tool_call_id"),
-        reasoning_content=msg.get("reasoning_content"),
+        reasoning_content=reasoning_value,
+        reasoning=reasoning_value,
     )
 
 
diff --git a/src/inference_endpoint/openai/types.py b/src/inference_endpoint/openai/types.py
index 400baf704..77fa16f2f 100644
--- a/src/inference_endpoint/openai/types.py
+++ b/src/inference_endpoint/openai/types.py
@@ -101,6 +101,7 @@ class ChatMessage(
     tool_calls: list[dict[str, Any]] | None = None
     tool_call_id: str | None = None
     reasoning_content: str | None = None
+    reasoning: str | None = None
 
 
 # gc=False: audit 2026-05: request containers set at construction; frozen=True blocks field reassignment.
diff --git a/tests/unit/dataset_manager/test_agentic_inference_dataset.py b/tests/unit/dataset_manager/test_agentic_inference_dataset.py
index 9d0df6a01..7bcda39ab 100644
--- a/tests/unit/dataset_manager/test_agentic_inference_dataset.py
+++ b/tests/unit/dataset_manager/test_agentic_inference_dataset.py
@@ -973,6 +973,51 @@ def test_build_metadata_pre_built_messages_no_tools():
     assert msgs[2] == {"role": "user", "content": "C"}
 
 
+@pytest.mark.unit
+@pytest.mark.parametrize(
+    ("message_fields", "expected_field"),
+    [
+        ({"reasoning_content": "thinking"}, "reasoning_content"),
+        ({"reasoning": "thinking"}, "reasoning"),
+        ({"reasoning_content": "thinking", "reasoning": "thinking"}, None),
+    ],
+    ids=["reasoning_content", "reasoning", "both_rejected"],
+)
+def test_pre_built_messages_reasoning_alias_policy(
+    message_fields: dict[str, str], expected_field: str | None
+):
+    df = pd.DataFrame(
+        [
+            {"conversation_id": "c1", "turn": 1, "role": "user", "content": "A"},
+            {
+                "conversation_id": "c1",
+                "turn": 2,
+                "role": "assistant",
+                "content": "B",
+                **message_fields,
+            },
+            {"conversation_id": "c1", "turn": 3, "role": "user", "content": "C"},
+        ]
+    )
+    ds = AgenticInferenceDataset(df)
+
+    if expected_field is None:
+        with pytest.raises(
+            InputValidationError,
+            match=(
+                "conversation 'c1' turn 2 has both "
+                "'reasoning_content' and 'reasoning'"
+            ),
+        ):
+            ds.load()
+        return
+
+    ds.load()
+    assert ds.conversation_metadata is not None
+    msgs = ds.conversation_metadata.pre_built_messages_by_key[("c1", 3)]
+    assert msgs[1][expected_field] == "thinking"
+
+
 @pytest.mark.unit
 def test_load_sample_includes_messages():
     """load_sample returns messages with the complete message list."""
diff --git a/tests/unit/openai/test_msgspec_adapter.py b/tests/unit/openai/test_msgspec_adapter.py
index 1fac94d26..a11c7aa52 100644
--- a/tests/unit/openai/test_msgspec_adapter.py
+++ b/tests/unit/openai/test_msgspec_adapter.py
@@ -165,6 +165,39 @@ def test_chat_message_from_dict_all_fields():
     assert msg.tool_call_id is None
 
 
+@pytest.mark.unit
+@pytest.mark.parametrize(
+    ("message_fields", "expected"),
+    [
+        ({"reasoning_content": "thinking via sglang"}, "thinking via sglang"),
+        ({"reasoning": "thinking via vllm"}, "thinking via vllm"),
+        ({"reasoning_content": "sglang", "reasoning": "vllm"}, "sglang"),
+    ],
+    ids=["reasoning_content", "reasoning", "reasoning_content_precedence"],
+)
+def test_to_endpoint_request_forwards_both_reasoning_aliases(message_fields, expected):
+    query = Query(
+        id="q-reasoning",
+        data={
+            "model": "m",
+            "messages": [
+                {
+                    "role": "assistant",
+                    "content": "answer",
+                    **message_fields,
+                }
+            ],
+        },
+    )
+
+    request = OpenAIMsgspecAdapter.to_endpoint_request(query)
+    payload = json.loads(msgspec.json.encode(request))
+
+    msg = payload["messages"][0]
+    assert msg["reasoning_content"] == expected
+    assert msg["reasoning"] == expected
+
+
 @pytest.mark.unit
 def test_chat_message_content_optional():
     """ChatMessage accepts content=None for tool-dispatching assistant turns."""