diff --git a/contributing/samples/gepa/experiment.py b/contributing/samples/gepa/experiment.py
index f3751206a8..2710c3894c 100644
--- a/contributing/samples/gepa/experiment.py
+++ b/contributing/samples/gepa/experiment.py
@@ -43,7 +43,6 @@
 from tau_bench.types import EnvRunResult
 from tau_bench.types import RunConfig
 import tau_bench_agent as tau_bench_agent_lib
-
 import utils
 
 
diff --git a/contributing/samples/gepa/run_experiment.py b/contributing/samples/gepa/run_experiment.py
index d857da9635..e31db15788 100644
--- a/contributing/samples/gepa/run_experiment.py
+++ b/contributing/samples/gepa/run_experiment.py
@@ -25,7 +25,6 @@
 from absl import flags
 import experiment
 from google.genai import types
-
 import utils
 
 _OUTPUT_DIR = flags.DEFINE_string(
diff --git a/src/google/adk/models/lite_llm.py b/src/google/adk/models/lite_llm.py
index b954d8a01f..9df436ded2 100644
--- a/src/google/adk/models/lite_llm.py
+++ b/src/google/adk/models/lite_llm.py
@@ -1387,13 +1387,7 @@ def _model_response_to_generate_content_response(
   if finish_reason:
     # If LiteLLM already provides a FinishReason enum (e.g., for Gemini), use
     # it directly. Otherwise, map the finish_reason string to the enum.
-    if isinstance(finish_reason, types.FinishReason):
-      llm_response.finish_reason = finish_reason
-    else:
-      finish_reason_str = str(finish_reason).lower()
-      llm_response.finish_reason = _FINISH_REASON_MAPPING.get(
-          finish_reason_str, types.FinishReason.OTHER
-      )
+    llm_response.finish_reason = _map_finish_reason(finish_reason)
   if response.get("usage", None):
     llm_response.usage_metadata = types.GenerateContentResponseUsageMetadata(
         prompt_token_count=response["usage"].get("prompt_tokens", 0),
@@ -1968,18 +1962,22 @@ async def generate_content_async(
                         ),
                     )
                 )
-            aggregated_llm_response_with_tool_call = (
-                _message_to_generate_content_response(
-                    ChatCompletionAssistantMessage(
-                        role="assistant",
-                        content=text,
-                        tool_calls=tool_calls,
-                    ),
-                    model_version=part.model,
-                    thought_parts=list(reasoning_parts)
-                    if reasoning_parts
-                    else None,
-                )
+            aggregated_llm_response_with_tool_call = _message_to_generate_content_response(
+                ChatCompletionAssistantMessage(
+                    role="assistant",
+                    # FIX: Set content=None for tool-only messages to avoid duplication
+                    # and follow OpenAI/LiteLLM conventions. Planning/reasoning text is
+                    # already streamed (lines 1288-1296) and preserved in thought_parts
+                    # (line 1357). Including it again in content causes duplication and
+                    # violates API specifications for tool-call messages.
+                    # See: https://github.com/google/adk-python/issues/3697
+                    content=None,
+                    tool_calls=tool_calls,
+                ),
+                model_version=part.model,
+                thought_parts=list(reasoning_parts)
+                if reasoning_parts
+                else None,
             )
             aggregated_llm_response_with_tool_call.finish_reason = (
                 _map_finish_reason(finish_reason)