Prove the lift · AI model performance evaluation

gemini-2.5-pro vs kimi-k2-thinking

-2-point lift for gemini-2.5-pro on insurance-extraction-eval

Test summary

This benchmark evaluates gemini-2.5-pro against kimi-k2-thinking on the insurance-extraction-eval v2 dataset (6 rows), scored by an LLM-as-judge (llama-3.1-8b). gemini-2.5-pro reached a 54% pass rate versus 56% for kimi-k2-thinking — a -2-point lift.

Base model
kimi-k2-thinking
via moonshot · 56% pass rate
Refined model
gemini-2.5-pro
via google · 54% pass rate · 0.8s p95
Base pass rate
56%
Refined pass rate
54%
Refined p95
0.8s
Est. tokens saved
1.4K
0pts
no lift — the base model held its ground
0%
Base model
0%
Invoked-refined
2 improved1 regressed3 unchanged

Per-row results

1Business-interruption period-of-restoration calculation0%100%+100
2Named-peril coverage determination under a basic HO form0%100%+100
3Aggregate deductible across a loss sequence under a per-occurrence + annual cap100%100%0
4Pro-rata distribution across layered carriers on a covered loss100%0%-100
5Coinsurance 80% clause recovery on an under-insured building0%0%0
6Replacement cost vs ACV on a depreciated roof100%100%0
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