[BUG] Switching to Sonnet 4.5 from Gemini 3.0 Pro mid-conversation causes an API error

[MichaelYochpaz]

Problem (one or two sentences)

When I start a conversation with (or switch mid-conversation to) Gemini 3.0, and then try to switch again to a different mode, that uses a different model, I get the following error for any message I try to send a message to Roo in the same conversation:

400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.37.content.1: Input tag 'thoughtSignature' found using 'type' does not match any of the expected tags: 'document', 'image', 'redacted_thinking', 'search_result', 'text', 'thinking', 'tool_result', 'tool_use'"},"request_id":"req_vrtx_011CVHNpz4TTYSzMjAZqCCgR"}

And then because of that, I'm forced to use Gemini 3 for the rest of the conversation, or start a new one.

I suspect the issue is because Gemini's endpoint uses specific API parameters that other endpoints, like Claude's, don't support. Then when switching from Gemini, it doesn't update the API parameters, and Roo Code attempts to communicate with other endpoints like Claude's, using Gemini-specific parameters.
This is just an unbased gut-feeling, so I might be completely wrong.

For additional context, if relevant, Gemini 3.0 Pro is configured using a "Google Gemini" provider, that's a Gemini API key generated on a GCP project (using Vertex AI), and Claude Sonnet 4.5 is configured using the "GCP Vertex AI" provider (on a different GCP project).

Context (who is affected and when)

When using Gemini 3, and then switching to use a different model in the same conversation.

Reproduction steps

1. Start a conversation with Roo using Gemini 3.0 Pro.
2. Switch to a different mode using a different model, and try to continue the conversation.
3. An error, similar to the one I got, should appear.

Expected result

I get a valid response

Actual result

Getting an error

Variations tried (optional)

No response

App Version

v3.33.1

API Provider (optional)

Google Vertex AI

Model Used (optional)

Gemini 3.0 Pro and Claude Sonnet 4.5

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I'm exploring a potential fix for this issue where switching from Gemini 3.0 to other models mid-conversation causes API errors due to incompatible message content types. I'll investigate the message transformation logic and open a PR to share progress.

[roomote]

Opened a PR with my attempt at addressing this issue: #9394

The fix filters out the Gemini-specific thoughtSignature blocks from conversation history when switching to other models, which should prevent the API validation errors you were experiencing.

Feedback welcome!

[hannesrudolph]

This was fixed in 3.33.1.

Just tested with exact repro steps and was unable to repro.

[candalfigomoro]

@hannesrudolph Still an issue with version 3.33.3.

1. Start a new conversation with Gemini 3.0 Pro.
2. Switch to Claude 4.5 Sonnet (I keep the same Code mode).
3. Send a new request in the same conversation.
4. ERROR (see attached screenshot).

400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.1.content.1: Input tag 'thoughtSignature' found using 'type' does not match any of the expected tags: 'document', 'image', 'redacted_thinking', 'search_result', 'server_tool_use', 'text', 'thinking', 'tool_result', 'tool_use', 'web_search_tool_result'"},"request_id":"