observability-llm-response-parsing-when-using-anthropic-reasoning-models-e-g-claude-h-a59c8c4a

THEN Modify the response extraction logic to iterate over the content array, concatenate all 'text' blocks into the assistant message, and optionally capture 'thinking' blocks as a separate field. Ensure the parser can handle 'thinking', 'text', 'tool_use', and 'tool_result' block types without breaking experiments or evaluators. This fix enables correct logging of reasoning models' outputs.

def parse_anthropic_content(content):
    text_parts = []
    thinking_parts = []
    for block in content:
        if block['type'] == 'text':
            text_parts.append(block['text'])
        elif block['type'] == 'thinking':
            thinking_parts.append(block['thinking'])
    return ' '.join(text_parts), ' '.join(thinking_parts)