High-reasoning mode configures the underlying Gemini model with thinking_level="high" and an expanded token budget to deliberately reason through natural placement strategies before outputting the draft.