Prove the lift · AI model performance evaluation

o4-mini vs claude-3-5-haiku

+21-point lift for o4-mini on insurance-extraction-eval

Test summary

This benchmark evaluates o4-mini against claude-3-5-haiku on the insurance-extraction-eval v2 dataset (8 rows), scored by an LLM-as-judge (claude-haiku-4-5). o4-mini reached a 56% pass rate versus 35% for claude-3-5-haiku — a +21-point lift.

Base model
claude-3-5-haiku
via anthropic · 35% pass rate
Refined model
o4-mini
via openai · 56% pass rate · 2.1s p95
Base pass rate
35%
Refined pass rate
56%
Refined p95
2.1s
Est. tokens saved
930
+0pts
refined beats base on 8 rows
0%
Base model
0%
Invoked-refined
6 improved0 regressed2 unchanged

Per-row results

1Named-peril coverage determination under a basic HO form100%100%0
2Aggregate deductible across a loss sequence under a per-occurrence + annual cap0%100%+100
3Pro-rata distribution across layered carriers on a covered loss0%100%+100
4Business-interruption period-of-restoration calculation0%100%+100
5Subrogation waiver interaction with additional-insured status0%100%+100
6Coinsurance 80% clause recovery on an under-insured building0%100%+100
7Suit-limitation clause: is a late-filed claim contractually barred?100%100%0
8Replacement cost vs ACV on a depreciated roof0%100%+100
Prove the lift on your own data.

Run any two models over your tasks, judged automatically, and see the improvement — in seconds.

Try Invoked
Generated with Invoked · Browse all benchmarks