Add support for Gemini 3 Pro on VertexAI

[alejandroem1]

Package (Required)

• langchain-google-genai
• langchain-google-vertexai
• langchain-google-community
• Other / not sure / general

Checked other resources

• I added a descriptive title to this issue
• I searched the LangChain documentation and API reference (linked above)
• I used the GitHub search to find a similar issue and didn't find it
• I am sure this is a feature request and not a bug report or question

Feature Description

The possibility of using Gemini 3 Pro with vertex AI in langchain

Use Case

It is important because our deployments are in Vertex

Proposed Solution

Alternatives Considered

No response

Additional Context

No response

[sradu]

+1

[borrna]

Same here, get NotFound: 404 Publisher Model when I try to use 'gemini-3-pro-preview' on VertexAI through langchain

[thderoo]

Same here, get NotFound: 404 Publisher Model when I try to use 'gemini-3-pro-preview' on VertexAI through langchain

Just set location='global'

[borrna]

Same here, get NotFound: 404 Publisher Model when I try to use 'gemini-3-pro-preview' on VertexAI through langchain

Just set location='global'

Hey, I just tried this and got MethodNotImplemented: 501 Received http2 header with status: 404

Full log: langchain_core.language_models.llms - WARNING - Retrying langchain_google_vertexai.chat_models._completion_with_retry.<locals>._completion_with_retry_inner in 4.0 seconds as it raised MethodNotImplemented: 501 Received http2 header with status: 404

[alejandroem1]

For us it is working but the problem is that the langchain vertexai package is lacking real support for the thinking_level parameter

[carlosmhe]

For us it is working but the problem is that the langchain vertexai package is lacking real support for the thinking_level parameter

On top of this, function calling is not working because there is no support for thought signatures. See Thought signatures

When using Gemini 3 Pro, you must pass back thought signatures during function calling, otherwise you will get a validation error (4xx status code).

[joris-swapfiets]

I've got the same issue with tool calling. Its not working because of no support for thought signatures. I'm using latest version of langchain-google-genai.

[thderoo]

I've got the same issue with tool calling. Its not working because of no support for thought signatures. I'm using latest version of langchain-google-genai.

You can get around it by providing fake thought signatures to the assistant messages containing tool calls (like it's done in #1365, there is an example of DUMMY_THOUGHT_SIGNATURE in libs/genai/langchain_google_genai/chat_models.py)

assistant_message_with_tool_calls['additional_kwargs']['__gemini_function_call_thought_signatures__'] = {
  '<tool call id 1>': DUMMY_THOUGHT_SIGNATURE,
  '<tool call id 2>': DUMMY_THOUGHT_SIGNATURE
  ...
}

But yeah a fix would be nice.

[lkuligin]

@carlosmhe could you provide a failing example, please? I tested a simple ReACT agent with Gemini-3 through ChatVertexAI and it's working fine on my side

[thderoo]

@carlosmhe could you provide a failing example, please? I tested a simple ReACT agent with Gemini-3 through ChatVertexAI and it's working fine on my side

It fails everytime there is a tool call without the corresponding thought signatures :

context = [
  { 'role': 'user', 'content': 'tool test' },
  { 'role': 'assistant', 'content': '', 'tool_calls': [ { 'name': 'test_tool', 'args': {}, 'id': 'call_1' } ] },
  { 'tool': 'tool', 'content': 'Success', 'name': 'test_tool', 'tool_call_id': 'call_1' }
]

model.invoke(context)

Maybe it works on your side because you don't need to simulate tool calls and all your assistant messages do contain thought signatures.

[thderoo]

Btw there is also a problem with neither thinking_budget or thinking_level being available on gemini 3 on vertex, I think it's probably more urgent than the rest because it cannot be easily overriden afaik.

[joris-swapfiets]

When I do parallel tool calls within a react agent using gemini 3.0 pro I get the following error:

langchain_google_genai.chat_models.ChatGoogleGenerativeAIError: Invalid argument provided to Gemini: 400 Function call is
 missing a thought_signature in functionCall parts. This is required for tools to work correctly, and missing thought_signature
 may lead to degraded model performance. Additional data, function call `default_api:<tool_name>` ,
 position 2. please refer to https://ai.google.dev/gemini-api/docs/thought-signatures for more details.