Vertex AI API - "gemini-3-pro-preview" - Support?

[JWebDev]

Hi,

I would like to try “gemini-3-pro-preview” with Vertex AI SDK. As far as I understand, support for “gemini-3-pro-preview” has not yet been implemented.
Please let me know when support for “gemini-3-pro-preview” is planned to be implemented for spring-ai-starter-model-vertex-ai-gemini?

I have updated to the latest version, but as far as I understand, it has not been implemented yet.

spring-ai-starter-model-vertex-ai-gemini => 1.1.1
ChatModel - I don't see any support yet.
Also, when I try to make an API call, I get a 404 error.

Thanks for the answer

[rking-bennie]

I am still getting the thinking config error when using gemini 3 as well. I notice the docs still don't say Gemini 3 is supported so I guess we have to wait :(

[rking-bennie]

To provide more context, if the MemoryChatAdvisor, ToolCallingAdvisor, and AugmentedToolCallbackProvider from the documentation here: is used, you will still receive thinking error as if augmented tool is not correctly providing context to gemini. This leads me to believe this only works with 2.5 because they are optional parameters. Gemini 3.0 made them required and that's where the issue shows up so the AugmentedToolCallbackProvider may be the issue. @JWebDev did you ever see anything different?

Caused by: io.grpc.StatusRuntimeException: INVALID_ARGUMENT: Unable to submit request because function call `retrieve_big_query_schema` in the 2. content block is missing a `thought_signature`. Learn more: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/thought-signatures
	at io.grpc.Status.asRuntimeException(Status.java:532)
	... 18 common frames omitted

[marcjaner]

Hi @JWebDev and @rking-bennie,

The root cause: spring-ai-starter-model-vertex-ai-gemini has no Gemini 3 support in Spring AI 1.x. Its model enum tops out at Gemini 2.5, with no ThinkingLevel or thought signature handling. The 404 and missing thought_signature errors both follow from this.

Gemini 3 support lives in spring-ai-starter-model-google-genai (available since 1.1.0), which handles thought signature passback automatically in the tool execution loop.

// remove
implementation 'org.springframework.ai:spring-ai-starter-model-vertex-ai-gemini'
// add
implementation 'org.springframework.ai:spring-ai-starter-model-google-genai'

spring.ai.google.genai.api-key=YOUR_API_KEY
# gemini-3-pro-preview was discontinued — use the current model ID
spring.ai.google.genai.chat.options.model=gemini-3.1-pro-preview

Leave internalToolExecutionEnabled at its default (true). Thought signatures are extracted and reattached automatically.

On Spring AI 2.0: spring-ai-starter-model-vertex-ai-gemini is formally deprecated as of 2.0.0-M4 with no planned Gemini 3 support. spring-ai-starter-model-google-genai is the forward path there too.

[rking-bennie]

@marcjaner By chance, are you using your own repository layer to save this or are you using the spring provided one? The reason I'm asking is because we needed to save the data in our own "shape" so I implemented my own ChatMemoryRepository. The small problem I'm having now is the thought signature comes in the metadata as a List<byte[]> but I am forced into persisting this to the db to return it and the problem is it's a List coming back from the DB so that causes a class cast exception within the Gemini model. I'm going to handle this in a special way for this metadata key but I was wondering if there was a plug-in-play solution for dealing with the metadata <String, Object> or else I'll need to be aware each time Google introduces a new key,