Prove the lift · AI model performance evaluation

claude-sonnet-4-6 vs grok-3

+19-point lift for claude-sonnet-4-6 on legal-research-qa

Test summary

This benchmark evaluates claude-sonnet-4-6 against grok-3 on the legal-research-qa v3 dataset (5 rows), scored by an LLM-as-judge (gpt-4o-mini, run locally). claude-sonnet-4-6 reached a 78% pass rate versus 59% for grok-3 — a +19-point lift.

Base model
grok-3
via xai · 59% pass rate
Refined model
claude-sonnet-4-6
via anthropic · 78% pass rate · 3.5s p95
Base pass rate
59%
Refined pass rate
78%
Refined p95
3.5s
Est. tokens saved
6.8K
+0pts
refined beats base on 5 rows
0%
Base model
0%
Invoked-refined
0 improved0 regressed4 unchanged

Per-row results

1Summarize the holding of a cited appellate case100%100%0
2Determine whether a precedent was overruled100%100%0
3Identify the controlling statute in a fact pattern100%100%0
4Extract the standard of review applied100%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