Gemini 3.8 Flash: thinking-token limits and static-video CDPA coverage

Google Cloud Build with AI Generative AI & Foundational Models
1

We are evaluating paid Google Cloud gemini-3.8-flash through the US multi-region v1 GenerateContent endpoint.

Proposed input: private same-project GCS MP4, approximately 22.3 MB and 130 seconds, H.264/AAC, default resolution, 2 FPS.

Settings: LOW thinking, includeThoughts=false, maxOutputTokens=16384, JSON responseSchema, no tools, grounding or explicit cache.

Could a Google representative clarify two documentation questions?

  1. Does maxOutputTokens include billable thinking tokens? What hard limits apply to thinking, visible output and their combined total? Does the published 65,536 maximum include thinking?

  2. The model is GA, but the video documentation labels Gemini 3 video tokenization Preview. Is default-resolution static v1 video processing covered by the Cloud Data Processing Addendum for identifiable people and incidental audio? The separate agentic-video section explicitly permits personal-data processing; does comparable coverage apply to this static path?

Please link the applicable official specifications or terms. No real footage has been submitted.