ai-agents-streaming-reasoning-when-streaming-from-gemini-gemini-2-5-flash-previe-3be192cc

THEN Modify the chunk_parser in ModelResponseIterator to detect thought/reasoning chunks (e.g., when includeThoughts is true) and place them in a separate field like reasoning_content or thinking_blocks instead of appending to content, similar to how Claude handles thinking blocks.

# In litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini.ModelResponseIterator.chunk_parser
# Detect if a chunk part is a thought (e.g., when includeThoughts=True in Gemini request)
if part.get('text') and is_thought:
    delta['reasoning_content'] = part['text']
else:
    delta['content'] = part['text']