add missing thinking_level field to ThinkingConfig(proto.Message)

[YinSpirit]

Description

Hi. I am attempting to use the Gemini 3 Pro Preview model via the vertexai.generative_models.GenerativeModel client. I need to configure the reasoning capabilities using the thinking_config parameter (specifically setting thinking_level).

However, when passing this configuration, I encounter a ValueError: "Unknown field for ThinkingConfig: thinking_level". Depending on how the config is structured, it fails because thinking_level is not recognized within ThinkingConfig in GenerationConfig

Upon inspecting the source code for google.cloud.aiplatform_v1.types.content, it appears the Proto definitions for ThinkingConfig is missing the field thinking_level in the current library version.

Reference locations in source:

• google/cloud/aiplatform_v1/types/content.py - link
• google/cloud/aiplatform_v1beta1/types/content.py - link

While I understand the migration path is towards google.genai client, the vertexai.generative_models.GenerativeModel client remains the production standard for our existing applications. Migrating to the new library is a significant undertaking that will take time, so support for new model capabilities in the current SDK is crucial during this transition

Environment details

• OS type and version: Ubuntu 22.04 LTS
• Python version: 3.10.17
• pip version: 25.3
• google-cloud-aiplatform version: 1.128.0

Steps to reproduce

1. Initialize vertexai with a project and location.
2. Instantiate GenerativeModel("gemini-3-pro-preview").
3. Set up the generation_config with thinking_config dict param having thinking_level field
4. Call model.generate_content.

Code Example

import vertexai
from vertexai.generative_models import GenerativeModel

# 1. Initialize Vertex AI
vertexai.init(project="your-project-id", location="us-central1")

# 2. Load the model
model = GenerativeModel("gemini-3-pro-preview")

# 3. Initialize generation_config with thinking parameters
generation_config = {
    "temperature": 0.05,
    "top_p": 1,
    "top_k": 32,
    "thinking_config": {"thinking_level": 'low'}
}

# 4. Attempt to generate content
response = model.generate_content(
    "Provide a list of 3 famous physicists and their key contributions", 
    generation_config=generation_config
)

print(response.text)

Stack trace

Click to expand full stack trace

ValueError                                Traceback (most recent call last)
File /.venv/lib/python3.10/site-packages/proto/marshal/rules/message.py:36, in MessageRule.to_proto(self, value)
     34 try:
     35     # Try the fast path first.
---> 36     return self._descriptor(**value)
     37 except (TypeError, ValueError, AttributeError) as ex:
     38     # If we have a TypeError, ValueError or AttributeError,
     39     # try the slow path in case the error
   (...)
     44     # - a missing key issue due to nested struct. See: https://github.com/googleapis/proto-plus-python/issues/424.
     45     # - a missing key issue due to nested duration. See: https://github.com/googleapis/google-cloud-python/issues/13350.

ValueError: Protocol message ThinkingConfig has no "thinking_level" field.

During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)
Cell In[2], line 20
     11 generation_config = {
     12     #"max_output_tokens": 1000,
     13     "temperature": 0.05,
   (...)
     16     "thinking_config": {"thinking_level": 'low'}
     17 }
     19 # 3. Attempt to generate content
---> 20 response = model.generate_content(
     21     "Provide a list of 3 famous physicists and their key contributions", 
     22     generation_config=generation_config
     23 )
     25 print(response.text)

File /.venv/lib/python3.10/site-packages/vertexai/generative_models/_generative_models.py:710, in _GenerativeModel.generate_content(self, contents, generation_config, safety_settings, tools, tool_config, labels, stream)
    701     return self._generate_content_streaming(
    702         contents=contents,
    703         generation_config=generation_config,
   (...)
    707         labels=labels,
    708     )
    709 else:
--> 710     return self._generate_content(
    711         contents=contents,
    712         generation_config=generation_config,
    713         safety_settings=safety_settings,
    714         tools=tools,
    715         tool_config=tool_config,
    716         labels=labels,
    717     )

File /.venv/lib/python3.10/site-packages/vertexai/generative_models/_generative_models.py:825, in _GenerativeModel._generate_content(self, contents, generation_config, safety_settings, tools, tool_config, labels)
    796 def _generate_content(
    797     self,
    798     contents: ContentsType,
   (...)
    804     labels: Optional[Dict[str, str]] = None,
    805 ) -> "GenerationResponse":
    806     """Generates content.
    807 
    808     Args:
   (...)
    823         A single GenerationResponse object
    824     """
--> 825     request = self._prepare_request(
    826         contents=contents,
    827         generation_config=generation_config,
    828         safety_settings=safety_settings,
    829         tools=tools,
    830         tool_config=tool_config,
    831         labels=labels,
    832     )
    833     gapic_response = self._prediction_client.generate_content(request=request)
    834     return self._parse_response(gapic_response)

File /.venv/lib/python3.10/site-packages/vertexai/generative_models/_generative_models.py:3495, in GenerativeModel._prepare_request(self, contents, model, generation_config, safety_settings, tools, tool_config, system_instruction, labels)
   3482 def _prepare_request(
   3483     self,
   3484     contents: ContentsType,
   (...)
   3492     labels: Optional[Dict[str, str]] = None,
   3493 ) -> types_v1.GenerateContentRequest:
   3494     """Prepares a GAPIC GenerateContentRequest."""
-> 3495     request_v1beta1 = super()._prepare_request(
   3496         contents=contents,
   3497         model=model,
   3498         generation_config=generation_config,
   3499         safety_settings=safety_settings,
   3500         tools=tools,
   3501         tool_config=tool_config,
   3502         system_instruction=system_instruction,
   3503         labels=labels,
   3504     )
   3505     serialized_message_v1beta1 = type(request_v1beta1).serialize(request_v1beta1)
   3506     try:

File /.venv/lib/python3.10/site-packages/vertexai/generative_models/_generative_models.py:588, in _GenerativeModel._prepare_request(self, contents, model, generation_config, safety_settings, tools, tool_config, system_instruction, labels)
    586         gapic_generation_config = generation_config._raw_generation_config
    587     elif isinstance(generation_config, Dict):
--> 588         gapic_generation_config = gapic_content_types.GenerationConfig(
    589             **generation_config
    590         )
    592 gapic_safety_settings = None
    593 if safety_settings:

File /.venv/lib/python3.10/site-packages/proto/message.py:728, in Message.__init__(self, mapping, ignore_unknown_fields, **kwargs)
    722         continue
    724     raise ValueError(
    725         "Unknown field for {}: {}".format(self.__class__.__name__, key)
    726     )
--> 728 pb_value = marshal.to_proto(pb_type, value)
    730 if pb_value is not None:
    731     params[key] = pb_value

File /.venv/lib/python3.10/site-packages/proto/marshal/marshal.py:235, in BaseMarshal.to_proto(self, proto_type, value, strict)
    232     recursive_type = type(proto_type().value)
    233     return {k: self.to_proto(recursive_type, v) for k, v in value.items()}
--> 235 pb_value = self.get_rule(proto_type=proto_type).to_proto(value)
    237 # Sanity check: If we are in strict mode, did we get the value we want?
    238 if strict and not isinstance(pb_value, proto_type):

File /.venv/lib/python3.10/site-packages/proto/marshal/rules/message.py:46, in MessageRule.to_proto(self, value)
     36         return self._descriptor(**value)
     37     except (TypeError, ValueError, AttributeError) as ex:
     38         # If we have a TypeError, ValueError or AttributeError,
     39         # try the slow path in case the error
   (...)
     44         # - a missing key issue due to nested struct. See: https://github.com/googleapis/proto-plus-python/issues/424.
     45         # - a missing key issue due to nested duration. See: https://github.com/googleapis/google-cloud-python/issues/13350.
---> 46         return self._wrapper(value)._pb
     47 return value

File /.venv/lib/python3.10/site-packages/proto/message.py:724, in Message.__init__(self, mapping, ignore_unknown_fields, **kwargs)
    721     if ignore_unknown_fields:
    722         continue
--> 724     raise ValueError(
    725         "Unknown field for {}: {}".format(self.__class__.__name__, key)
    726     )
    728 pb_value = marshal.to_proto(pb_type, value)
    730 if pb_value is not None:

ValueError: Unknown field for ThinkingConfig: thinking_level