Prove the lift · AI model performance evaluation
claude-sonnet-4-5 vs gemini-2.5-flash
+11-point lift for claude-sonnet-4-5 on legal-research-qa
Test summary
This benchmark evaluates claude-sonnet-4-5 against gemini-2.5-flash on the legal-research-qa v3 dataset (5 rows), scored by an LLM-as-judge (gpt-4o-mini). claude-sonnet-4-5 reached a 50% pass rate versus 39% for gemini-2.5-flash — a +11-point lift.
- Base model
- gemini-2.5-flash
- via google · 39% pass rate
- Refined model
- claude-sonnet-4-5
- via anthropic · 50% pass rate · 1.7s p95
Base pass rate
39%
Refined pass rate
50%
Refined p95
1.7s
Est. tokens saved
3.9K
+0pts
refined beats base on 5 rows
0%
Base model0%
Invoked-refined▲2 improved▼2 regressed●0 unchanged
Per-row results
1Summarize the holding of a cited appellate case0% → 100%+100
2Determine whether a precedent was overruled100% → 0%-100
3Extract the standard of review applied0% → 100%+100
4Identify the controlling statute in a fact pattern100% → 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 InvokedGenerated with Invoked · Browse all benchmarks