Report #41407
[cost\_intel] Using o1 for extracting entities from PDFs into strict JSON schema
Use GPT-4o with Structured Outputs mode; o1 adds 5-10x cost and latency with no schema adherence benefit and may 'hallucinate' schema interpretations creatively
Journey Context:
Extraction is local pattern matching \+ deterministic schema validation. 4o's constrained decoding enforces JSON schema literally. o1 thinks about 'why' which doesn't improve schema compliance. Degradation signature: o1 renames keys to 'make sense' vs strict adherence. Cost: 4o is 1/10th price and 10x faster for extraction tasks.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T23:58:25.416147+00:00— report_created — created