diff --git a/components/src/dynamo/frontend/prepost.py b/components/src/dynamo/frontend/prepost.py
index 003813e99ce6..0b92fdcc6cb0 100644
--- a/components/src/dynamo/frontend/prepost.py
+++ b/components/src/dynamo/frontend/prepost.py
@@ -136,8 +136,19 @@ def _prepare_request(
         )
         else None
     )
-    chat_template_kwargs = dict(request_for_sampling.chat_template_kwargs or {})
-    chat_template_kwargs["reasoning_effort"] = request_for_sampling.reasoning_effort
+    # serde's `alias` is deserialize-only, so pythonize emits the Rust field
+    # name `chat_template_args`; read it too or client kwargs are dropped.
+    raw_template_args = (
+        request.get("chat_template_args") if isinstance(request, dict) else None
+    )
+    chat_template_kwargs = dict(
+        request_for_sampling.chat_template_kwargs or raw_template_args or {}
+    )
+    # Don't let an absent top-level field clobber a nested reasoning_effort.
+    if request_for_sampling.reasoning_effort is not None:
+        chat_template_kwargs["reasoning_effort"] = request_for_sampling.reasoning_effort
+    else:
+        chat_template_kwargs.setdefault("reasoning_effort", None)
 
     # Mistral warns that tokenize=False is unsafe for chat templates.
     is_mistral_tokenizer = (
@@ -154,14 +165,15 @@ def _prepare_request(
     chat_params = ChatParams(
         chat_template=request_for_sampling.chat_template,
         chat_template_content_format="auto",
-        chat_template_kwargs=dict(
-            add_generation_prompt=request_for_sampling.add_generation_prompt,
-            continue_final_message=request_for_sampling.continue_final_message,
-            tools=tool_dicts,
-            documents=request_for_sampling.documents,
-            tokenize=tokenize_in_template,
+        # Renderer-managed keys last so a nested duplicate can't raise TypeError.
+        chat_template_kwargs={
             **chat_template_kwargs,
-        ),
+            "add_generation_prompt": request_for_sampling.add_generation_prompt,
+            "continue_final_message": request_for_sampling.continue_final_message,
+            "tools": tool_dicts,
+            "documents": request_for_sampling.documents,
+            "tokenize": tokenize_in_template,
+        },
     )
 
     return (
diff --git a/components/src/dynamo/frontend/tests/test_vllm_processor_unit.py b/components/src/dynamo/frontend/tests/test_vllm_processor_unit.py
index ed54cbe82d9b..5ebdc355edb7 100644
--- a/components/src/dynamo/frontend/tests/test_vllm_processor_unit.py
+++ b/components/src/dynamo/frontend/tests/test_vllm_processor_unit.py
@@ -155,6 +155,88 @@ def test_no_tools_in_request(self, tokenizer):
         ), "No tools in request should produce None tools in template"
 
 
+class TestChatTemplateArgsPassthrough:
+    """Per-request chat template kwargs must survive into the rendered template.
+
+    pythonize serializes the request field under its Rust name
+    ``chat_template_args`` (serde ``alias`` is deserialize-only), so the vLLM
+    processor must read that key, not only vLLM's native ``chat_template_kwargs``.
+    """
+
+    def test_chat_template_args_reaches_template(self, tokenizer):
+        """Kwargs keyed as chat_template_args (the pythonize'd key) reach the template."""
+        _, _, _, _, chat_params = _prepare_request(
+            {
+                "model": MODEL,
+                "messages": [{"role": "user", "content": "Hello"}],
+                "chat_template_args": {"enable_thinking": False},
+            },
+            tokenizer=tokenizer,
+            tool_parser_class=None,
+        )
+        assert (
+            chat_params.chat_template_kwargs.get("enable_thinking") is False
+        ), "chat_template_args must be forwarded to the chat template"
+
+    def test_chat_template_kwargs_native_key_still_works(self, tokenizer):
+        """The vLLM-native chat_template_kwargs key keeps working."""
+        _, _, _, _, chat_params = _prepare_request(
+            {
+                "model": MODEL,
+                "messages": [{"role": "user", "content": "Hello"}],
+                "chat_template_kwargs": {"enable_thinking": False},
+            },
+            tokenizer=tokenizer,
+            tool_parser_class=None,
+        )
+        assert (
+            chat_params.chat_template_kwargs.get("enable_thinking") is False
+        ), "native chat_template_kwargs must be forwarded to the chat template"
+
+    def test_nested_reasoning_effort_is_not_clobbered(self, tokenizer):
+        """A reasoning_effort nested in template kwargs survives the top-level default."""
+        _, _, _, _, chat_params = _prepare_request(
+            {
+                "model": MODEL,
+                "messages": [{"role": "user", "content": "Hello"}],
+                "chat_template_args": {"reasoning_effort": "high"},
+            },
+            tokenizer=tokenizer,
+            tool_parser_class=None,
+        )
+        assert (
+            chat_params.chat_template_kwargs.get("reasoning_effort") == "high"
+        ), "nested reasoning_effort must not be overwritten by an absent top-level field"
+
+    def test_top_level_reasoning_effort_wins_over_nested(self, tokenizer):
+        """An explicit top-level reasoning_effort overrides a nested one."""
+        _, _, _, _, chat_params = _prepare_request(
+            {
+                "model": MODEL,
+                "messages": [{"role": "user", "content": "Hello"}],
+                "reasoning_effort": "low",
+                "chat_template_args": {"reasoning_effort": "high"},
+            },
+            tokenizer=tokenizer,
+            tool_parser_class=None,
+        )
+        assert chat_params.chat_template_kwargs.get("reasoning_effort") == "low"
+
+    def test_reserved_render_key_in_template_args_does_not_crash(self, tokenizer):
+        """A renderer-reserved key nested in template kwargs must not raise TypeError."""
+        _, _, _, _, chat_params = _prepare_request(
+            {
+                "model": MODEL,
+                "messages": [{"role": "user", "content": "Hello"}],
+                "chat_template_args": {"documents": [{"text": "doc"}]},
+            },
+            tokenizer=tokenizer,
+            tool_parser_class=None,
+        )
+        # Renderer-managed value wins over the client's nested key (no crash either).
+        assert chat_params.chat_template_kwargs["documents"] is None
+
+
 class TestMultimodalFeatureMetadata:
     def _feature(
         self, modality, mm_hash, offset, length, data=_DEFAULT_MM_DATA, is_embed=None