We’re finding that intermittently, but increasingly, our gemini 2.5 pro model results are returning output which does not contain any thoughts. It shows it has used some of the thinking budget (thoughtsTokenCount), but will just return the text output without its reasoning. Anyone else running into this issue, have you resolved it?
The config uses:
"thinkingConfig": {
"includeThoughts": true,
"thinkingBudget": 20000
}