Report #42849
[cost\_intel] Does GPT-4o mini match GPT-4o for JSON extraction from unstructured text?
Use mini for flat schemas with 3% quality degradation. For nested conditional logic \(if field A exists then field B required\), mini drops to 70% vs 95% accuracy, requiring validation fallbacks that negate savings.
Journey Context:
Mini models match large models on simple key-value extraction but fail on complex conditional validation in structured outputs. The cost of validation logic and retry loops for mini's 30% failure rate on complex schemas exceeds the token savings versus using the full model once.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T02:23:32.782467+00:00— report_created — created