diff --git a/src/google/adk/flows/llm_flows/contents.py b/src/google/adk/flows/llm_flows/contents.py
index 62728124b81..a4aeba56c4d 100644
--- a/src/google/adk/flows/llm_flows/contents.py
+++ b/src/google/adk/flows/llm_flows/contents.py
@@ -386,7 +386,7 @@ def _build_task_input_user_content(
         try:
           import json as _json
 
-          text = _json.dumps(dict(fc.args))
+          text = _json.dumps(dict(fc.args), ensure_ascii=False)
         except (TypeError, ValueError):
           text = str(fc.args)
         parts = [types.Part(text=text)]
diff --git a/src/google/adk/utils/content_utils.py b/src/google/adk/utils/content_utils.py
index 9f6e8d4a9e6..d0bada24015 100644
--- a/src/google/adk/utils/content_utils.py
+++ b/src/google/adk/utils/content_utils.py
@@ -64,7 +64,7 @@ def to_user_content(value: Any) -> types.Content:
     deep-copied)
   - str -> single text part
   - BaseModel -> model_dump_json() text part
-  - dict/list -> json.dumps() text part
+  - dict/list -> json.dumps() text part (non-ASCII preserved, not escaped)
   - anything else -> str() text part
   """
   if isinstance(value, types.Content):
@@ -74,7 +74,7 @@ def to_user_content(value: Any) -> types.Content:
   elif isinstance(value, BaseModel):
     text = value.model_dump_json()
   elif isinstance(value, (dict, list)):
-    text = json.dumps(value)
+    text = json.dumps(value, ensure_ascii=False)
   else:
     text = str(value)
   return types.Content(role='user', parts=[types.Part(text=text)])
diff --git a/tests/unittests/flows/llm_flows/test_contents.py b/tests/unittests/flows/llm_flows/test_contents.py
index 930de059b77..479b279926c 100644
--- a/tests/unittests/flows/llm_flows/test_contents.py
+++ b/tests/unittests/flows/llm_flows/test_contents.py
@@ -1894,3 +1894,41 @@ def test_recover_compacted_parallel_call_reinjects_sibling_response():
       if part.function_response
   }
   assert response_ids == {"lr-1", "reg-1"}
+
+
+def test_task_input_user_content_preserves_non_ascii():
+  """Delegated task input must not escape non-ASCII FC args (issue #6279).
+
+  A chat coordinator delegates to a task sub-agent via a function call; the
+  task agent's first user turn is rebuilt from the FC args. Escaping non-Latin
+  characters to ``\\uXXXX`` there bloats prompt tokens and degrades responses.
+  """
+  fc_id = "fc_task_1"
+  events = [
+      Event(
+          invocation_id="inv1",
+          author="coordinator",
+          content=types.Content(
+              role="model",
+              parts=[
+                  types.Part(
+                      function_call=types.FunctionCall(
+                          id=fc_id,
+                          name="delegate",
+                          args={"query": "שלום עולם", "city": "北京"},
+                      )
+                  )
+              ],
+          ),
+      ),
+  ]
+
+  content = contents._build_task_input_user_content(  # pylint: disable=protected-access
+      events, isolation_scope=fc_id
+  )
+
+  assert content is not None and content.parts
+  text = content.parts[0].text
+  assert "שלום עולם" in text
+  assert "北京" in text
+  assert "\\u" not in text
diff --git a/tests/unittests/utils/test_content_utils.py b/tests/unittests/utils/test_content_utils.py
index b4a97e3c4a4..b9bad110af9 100644
--- a/tests/unittests/utils/test_content_utils.py
+++ b/tests/unittests/utils/test_content_utils.py
@@ -66,3 +66,25 @@ def test_to_user_content_other_input_is_str():
   content = to_user_content(42)
   assert content.role == 'user'
   assert content.parts[0].text == '42'
+
+
+def test_to_user_content_dict_input_preserves_non_ascii():
+  """Non-ASCII input must reach the LLM as-is, not as \\uXXXX escapes.
+
+  Escaping (json.dumps' default ensure_ascii=True) turns each non-Latin
+  character into a ``\\uXXXX`` sequence, which bloats prompt tokens and
+  degrades model responses for non-English inputs (issue #6279).
+  """
+  content = to_user_content({'query': 'שלום עולם', 'city': '北京'})
+  text = content.parts[0].text
+  assert 'שלום עולם' in text
+  assert '北京' in text
+  assert '\\u' not in text
+
+
+def test_to_user_content_list_input_preserves_non_ascii():
+  content = to_user_content(['שלום', '你好'])
+  text = content.parts[0].text
+  assert 'שלום' in text
+  assert '你好' in text
+  assert '\\u' not in text