Model thoughts stopped showing when using function calling

Gemini API
, ,
1

Since yesterday, model thought summary parts are not being returned when using function calling tool.

The “functionCall” part in the response does have a “thoughtSignature” field, however the other text part in the response is empty.

Up until yesterday the response contained 2 parts: the “functionCall” part and another text part with the field “thought“ set to “true”

I am using “gemini-3-flash-preview” with the following thinking config:
thinking_config = types.ThinkingConfig( include_thoughts=True, thinking_level=ThinkingLevel.HIGH )

1 Like
3

I’m experiencing the same issue since yesterday with gemini-3-flash-preview and gemini-3-pro-preview

Configuration:

thinkingConfig: {
  thinkingBudget: 2000,
  includeThoughts: true
}

What I observe:

  • usageMetadata.thoughtsTokenCount is returned with a number of token used
  • But no thought: true part in the response
  • Only thoughtSignature is present

Note: I’m not using function calling - this is a simple text generation request.

Switching to gemini-2.5-flash with the same configuration works correctly - I get the separate thought part as expected.

Using @google/genai SDK with TypeScript.

1 Like
4

looks like this is now fixed, I see the thoughts again :partying_face:

6

it look better, but not fully resolved on my side. I still see the issue in ~30% of requests.