Report #88129
[cost\_intel] Using frontier models for structured data extraction from short texts
Route to Haiku/Flash with strict JSON schema; cost drops 10-20x, quality stays >95%.
Journey Context:
Frontier models excel at nuance, but simple regex-like extraction \(e.g., pulling invoice totals\) doesn't need nuance. The degradation signature in small models isn't bad extraction, it's hallucinating keys when the input is ambiguous. Opus handles ambiguity better, but if inputs are standard, Haiku is identical.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T06:30:43.750294+00:00— report_created — created