Thought summaries not returned for Gemini 3.x (streamGenerateContent and Interactions API), even though thinking tokens are used

Gemini API
1

Hi, I’m not receiving any thought summaries from Gemini 3.x models, using either streamGenerateContent or the Interactions API. The models do think (thoughtsTokenCount / total_thought_tokens are non-zero), but no summary content is returned.

1. streamGenerateContent with includeThoughts: true

curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.5-flash:streamGenerateContent?alt=sse" \
  -H "x-goog-api-key: $GEMINI_API_KEY" -H "Content-Type: application/json" \
  -d '{
    "contents": [{"role": "user", "parts": [{"text": "Alice, Bob, and Carol each live in a different house on the same street: red, green, and blue. The person who lives in the red house owns a cat. Bob does not live in the green house. Carol owns a dog. The green house is to the left of the red house. Alice does not own a cat. Who lives in each house, and what pet do they own?"}]}],
    "generationConfig": {"thinkingConfig": {"includeThoughts": true}}
  }'
Model Thought tokens Thought parts received
gemini-3.8-flash 421 (thoughtsTokenCount) 0
gemini-3.8-flash, thinkingLevel: "high" (first answer token at ~9.5s) 0
gemini-3.5-flash (docs’ example prompt above, unchanged) 6570 0
gemini-3.5-flash-lite — 0

2. Interactions API with thinking_summaries: "auto"

curl -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" -H "Content-Type: application/json" --no-buffer \
  -d '{"model": "gemini-3.5-flash", "input": "How do I check login status with middleware in Next.js?", "generation_config": {"thinking_summaries": "auto"}, "stream": true}'
  • total_thought_tokens: 1325
  • Events received: step.start (type thought) → step.delta (thought_signature) → step.stop → model_output steps
  • The thought step is {"step":{"type":"thought"}} with no summary field, and no thought_summary delta is ever sent.

The docs say a summary may be empty for simple requests, but this happens even for the docs’ own logic puzzle with 6,570 thought tokens.

Environment

  • Date: 2026-10-09
  • API key created in Google AI Studio, called via REST (no SDK)
  • Tier: Free

Is this expected for these models, or is something wrong on my side? Thanks!