Prove the lift · AI model performance evaluation

llama-4-scout vs claude-3-5-haiku

+28-point lift for llama-4-scout on contract-clause-qa

Test summary

This benchmark evaluates llama-4-scout against claude-3-5-haiku on the contract-clause-qa v3 dataset (5 rows), scored by an LLM-as-judge (llama-3.1-8b). llama-4-scout reached a 72% pass rate versus 44% for claude-3-5-haiku — a +28-point lift.

Base model
claude-3-5-haiku
via anthropic · 44% pass rate
Refined model
llama-4-scout
via meta · 72% pass rate · 1.8s p95
Base pass rate
44%
Refined pass rate
72%
Refined p95
1.8s
Est. tokens saved
12.3K
+0pts
refined beats base on 5 rows
0%
Base model
0%
Invoked-refined
▲2 improved▼0 regressed●3 unchanged

Per-row results

1Extract the auto-renewal notice window in days100% → 100%0
2Identify the governing-law clause and jurisdiction0% → 100%+100
3Determine indemnification scope and carve-outs0% → 100%+100
4Find the liability cap and its exceptions100% → 100%0
5Locate the termination-for-convenience terms100% → 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