Report #90450
[cost\_intel] Using o1 for all NL2SQL generation regardless of schema complexity
Use GPT-4o for single-table queries; escalate to o1 only for >3 joins or nested window functions with complex business logic
Journey Context:
Spider benchmark shows 4o achieves 92% on easy/medium versus o1 94%; on hard \(nested queries\) 4o drops to 65% versus o1 88%; complexity threshold is visible at join depth >2. The cost cliff is 30x but only justified beyond complexity threshold.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T10:24:56.670612+00:00— report_created — created