Unable to obtain grounding data from response. using "gemini-3.1-flash-lite-preview"

[satounity]

(I don't know English so please excuse the machine translation.)

Grounding using Vertex AI Search in "gemini-3.1-flash-lite-preview".
No grounding data found in response.candidates[0].grounding_metadata.grounding_chunks.

[Example]
response = client.models.generate_content(...
...

print(response.candidates[0].grounding_metadata.grounding_chunks)
->No results

[satounity]

Add a note

――――――――――――――――――――
# model_id = "gemini-3-flash-preview"
model_id = "gemini-3.1-flash-lite-preview"
response = client.models.generate_content(
    model=model_id,
    contents=prompt_text,
    config=GenerateContentConfig(
        response_mime_type="application/json",
        # Grounding [Vertex AI Search]
        tools=[
            Tool(
                retrieval=Retrieval(
                    vertex_ai_search=VertexAISearch(
                        datastore=datastore,
                    )
                )
            )
        ],
        thinking_config=ThinkingConfig(
            thinking_level="LOW",
        ),
    ),
)
print(response.candidates[0].grounding_metadata.grounding_chunks)
――――――――――――――――――――

In the above code snippet,

●If the model is "gemini-3-flash-preview",
the
"print(response.candidates[0].grounding_metadata.grounding_chunks)"
command returns a result.

●If the model is "gemini-3.1-flash-lite-preview",
the
"print(response.candidates[0].grounding_metadata.grounding_chunks)"
command does not return a result.

[yk-kd]

The issue of code execution output files not being output to inlineData is probably related to this same problem.
I'm experiencing this when running with Node, so it might not be an SDK issue.

[eezdev]

I am experiencing the same issue using the C# / .NET SDK (Google.AI.GenerativeAI).

Even when the model performs multiple searches (e.g., grounding_searches_used: 10+) and the search_entry_point contains fully rendered HTML with valid source chips, the grounding_chunks array in the response metadata is frequently empty.

Observed Behavior:

Search Entry Point: Contains valid URIs and titles (rendered in the Google Search "chips").

Grounding Supports: Contains segments and grounding_chunk_indices, but these indices point to a null or empty grounding_chunks collection.

Text Content: The model generates text with citations/URLs, but because the metadata chunks are missing, there is no way to programmatically verify or map these back to the grounding_supports indices.

Impact:
This makes it impossible to build a reliable "Grounding".

Environment:

Language: C# / .NET

Package: Google.GenAI (NuGet)

Model: gemini-3xxxx (same with all emini-3x models)

This seems to be a cross-SDK issue (likely at the API/Protobuf level) where the grounding metadata is being stripped or not populated even when the search execution is successful.

[Venkaiahbabuneelam]

Hi @yk-kd
Thanks for reaching out us, we have tried to reproduce the issue but we were able to get the response results using both the models. Could you please let us know what is your environment configurations

I have used below environment configurations:

google genai: 1.66.0
VertexAI: True
model_id = gemini-3-flash-preview, gemini-3.1-flash-lite-preview

Thanks

[yk-kd]

Thank you for your reply, @Venkaiahbabuneelam .

I ran this script.
google genai: 1.66.0
VertexAI: True
model_id = gemini-3-flash-preview, gemini-3.1-flash-lite-preview

# /// script
# requires-python = ">=3.12"
# dependencies = [
#     "google-genai==1.66.0",
# ]
# ///

from google import genai
from google.genai import types

client = genai.Client(
    vertexai=True
)

SYSTEM_PROMPT = """\
You are a data processing assistant that uses Code Execution to process data.

Rules:
- Debug print() statements are allowed.
- You MUST write all final data processing results to a file. Do NOT print data results to stdout.
- Save result files to the working directory. They will be automatically returned as inlineData in the response.
"""

response = client.models.generate_content(
    model='gemini-3.1-flash-lite-preview',
    contents='Calculate the first 20 Fibonacci numbers and write them to a CSV file(output.csv)',
    config=types.GenerateContentConfig(
        system_instruction=SYSTEM_PROMPT,
        tools=[types.Tool(code_execution=types.ToolCodeExecution)]
    ),
)

import json
print("=== parts dump ===")
print(json.dumps([p.model_dump() for p in response.candidates[0].content.parts], indent=2, default=str))
print("==================")

Below are the results.

gemini-3.1-flash-lite-preview results

no inline_data

=== parts dump ===
[
  {
    "media_resolution": null,
    "code_execution_result": null,
    "executable_code": {
      "code": "import csv\n\ndef generate_fibonacci(n):\n    fib_sequence = []\n    a, b = 0, 1\n    for _ in range(n):\n        fib_sequence.append(a)\n        a, b = b, a + b\n    return fib_sequence\n\n# Calculate the first 20 Fibonacci numbers\nfib_20 = generate_fibonacci(20)\n\n# Write to CSV\nwith open('output.csv', mode='w', newline='') as file:\n    writer = csv.writer(file)\n    writer.writerow(['Index', 'Fibonacci Number'])\n    for i, num in enumerate(fib_20):\n        writer.writerow([i + 1, num])\n",
      "language": "PYTHON"
    },
    "file_data": null,
    "function_call": null,
    "function_response": null,
    "inline_data": null,
    "text": null,
    "thought": null,
    "thought_signature": "b'\\x01\\x8f=k_\\x17*\\xeb\\x12\\x1eiy\\xb2\\xd2\\xb7\\xc0Ds5\\xd5\\xda%\\xdc\\x8b\\xf8\\x97\\x17\\x1c*\\xbe\\xa3\\t\\x18\\x05\\x1e\\x16\\x15\\x01\\x00\\xd5\\xd3\\xecx{[\\xcc\\x95\\x02_\\xc5'",
    "video_metadata": null
  },
  {
    "media_resolution": null,
    "code_execution_result": {
      "outcome": "OUTCOME_OK",
      "output": "\n"
    },
    "executable_code": null,
    "file_data": null,
    "function_call": null,
    "function_response": null,
    "inline_data": null,
    "text": null,
    "thought": null,
    "thought_signature": null,
    "video_metadata": null
  },
  {
    "media_resolution": null,
    "code_execution_result": null,
    "executable_code": null,
    "file_data": null,
    "function_call": null,
    "function_response": null,
    "inline_data": null,
    "text": "The first 20 Fibonacci numbers have been calculated and saved to `output.csv`.",
    "thought": null,
    "thought_signature": "b'\\x01\\x8f=k_\\x1f\\xbd\\xab\\xee(,\\xf1\\xdf\\xaa\\x10[\\xbcC\\xd5+\\x9a\\xcc\\xeb\\x931RXv\\xd7q5\\x04,\\xd1\\xae\\x1c\\xf0\\x8c\\x9d\\x82%\\x81,\\x84}\\x80a:\\x12\\xad'",
    "video_metadata": null
  }
]
==================

gemini-3-flash-preview results

The third element of the array contains inline_data.

=== parts dump ===
[
  {
    "media_resolution": null,
    "code_execution_result": null,
    "executable_code": {
      "code": "import csv\n\ndef generate_fibonacci(n):\n    fib_sequence = []\n    a, b = 0, 1\n    for _ in range(n):\n        fib_sequence.append(a)\n        a, b = b, a + b\n    return fib_sequence\n\n# Generate the first 20 Fibonacci numbers\nfib_numbers = generate_fibonacci(20)\n\n# Write to CSV\nwith open('output.csv', mode='w', newline='') as file:\n    writer = csv.writer(file)\n    writer.writerow(['Index', 'Fibonacci Number'])\n    for index, value in enumerate(fib_numbers):\n        writer.writerow([index, value])\n\nprint(f\"Successfully generated {len(fib_numbers)} Fibonacci numbers and saved to output.csv\")\n",
      "language": "PYTHON"
    },
    "file_data": null,
    "function_call": null,
    "function_response": null,
    "inline_data": null,
    "text": null,
    "thought": null,
    "thought_signature": null,
    "video_metadata": null
  },
  {
    "media_resolution": null,
    "code_execution_result": {
      "outcome": "OUTCOME_OK",
      "output": "Successfully generated 20 Fibonacci numbers and saved to output.csv\n"
    },
    "executable_code": null,
    "file_data": null,
    "function_call": null,
    "function_response": null,
    "inline_data": null,
    "text": null,
    "thought": null,
    "thought_signature": null,
    "video_metadata": null
  },
  {
    "media_resolution": null,
    "code_execution_result": null,
    "executable_code": null,
    "file_data": null,
    "function_call": null,
    "function_response": null,
    "inline_data": {
      "data": "b'Index,Fibonacci Number\\r\\n0,0\\r\\n1,1\\r\\n2,1\\r\\n3,2\\r\\n4,3\\r\\n5,5\\r\\n6,8\\r\\n7,13\\r\\n8,21\\r\\n9,34\\r\\n10,55\\r\\n11,89\\r\\n12,144\\r\\n13,233\\r\\n14,377\\r\\n15,610\\r\\n16,987\\r\\n17,1597\\r\\n18,2584\\r\\n19,4181\\r\\n'",
      "display_name": null,
      "mime_type": "text/csv"
    },
    "text": null,
    "thought": null,
    "thought_signature": null,
    "video_metadata": null
  },
    ....
  }
]
==================