Prove the lift · AI model performance evaluation

claude-sonnet-4-6 vs llama-3.1-8b

-5-point lift for claude-sonnet-4-6 on insurance-extraction-eval

Test summary

This benchmark evaluates claude-sonnet-4-6 against llama-3.1-8b on the insurance-extraction-eval v1 dataset (8 rows), scored by an LLM-as-judge (claude-haiku-4-5). claude-sonnet-4-6 reached a 28% pass rate versus 33% for llama-3.1-8b — a -5-point lift.

Base model
llama-3.1-8b
via ollama · 33% pass rate
Refined model
claude-sonnet-4-6
via anthropic · 28% pass rate · 3.9s p95
Base pass rate
33%
Refined pass rate
28%
Refined p95
3.9s
Est. tokens saved
0pts
no lift — the base model held its ground
0%
Base model
0%
Invoked-refined
3 improved0 regressed5 unchanged

Per-row results

1Replacement cost vs ACV on a depreciated roof0%100%+100
2Coinsurance 80% clause recovery on an under-insured building0%0%0
3Aggregate deductible across a loss sequence under a per-occurrence + annual cap0%0%0
4Subrogation waiver interaction with additional-insured status0%100%+100
5Business-interruption period-of-restoration calculation100%100%0
6Pro-rata distribution across layered carriers on a covered loss0%0%0
7Suit-limitation clause: is a late-filed claim contractually barred?0%100%+100
8Named-peril coverage determination under a basic HO form0%0%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