Prove the lift · AI model performance evaluation

gpt-5 vs kimi-k2-thinking

+8-point lift for gpt-5 on insurance-extraction-eval

Test summary

This benchmark evaluates gpt-5 against kimi-k2-thinking on the insurance-extraction-eval v1 dataset (7 rows), scored by an LLM-as-judge (claude-haiku-4-5). gpt-5 reached a 66% pass rate versus 58% for kimi-k2-thinking — a +8-point lift.

Base model
kimi-k2-thinking
via moonshot · 58% pass rate
Refined model
gpt-5
via openai · 66% pass rate · 0.6s p95
Base pass rate
58%
Refined pass rate
66%
Refined p95
0.6s
Est. tokens saved
2.2K
+0pts
refined beats base on 7 rows
0%
Base model
0%
Invoked-refined
1 improved1 regressed5 unchanged

Per-row results

1Pro-rata distribution across layered carriers on a covered loss0%100%+100
2Suit-limitation clause: is a late-filed claim contractually barred?100%100%0
3Coinsurance 80% clause recovery on an under-insured building0%0%0
4Business-interruption period-of-restoration calculation100%100%0
5Replacement cost vs ACV on a depreciated roof100%100%0
6Subrogation waiver interaction with additional-insured status100%100%0
7Aggregate deductible across a loss sequence under a per-occurrence + annual cap100%0%-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