Langextract vs pydantic + API calls for no char_span use case

[karyotakisg]

Hi! I'm evaluating LangExtract for a pipeline that extracts structured entities from hundreds of 100+ page PDFs using LLMs + Pydantic models (currently via Instructor + OpenAI).

I don't need positional extraction (char_span / text anchoring). My main concerns are:

• Quality. Does LangExtract improve extraction accuracy over a straightforward Pydantic + LLM approach (e.g. Instructor)?
• Cost efficiency. Is langExtract a more expensive option?
• Speed. How does it handle mass extraction at scale?
  Or is text anchoring the primary differentiator, making it less relevant for my use case?