Prove the lift · AI model performance evaluation
gemini-2.5-flash vs grok-3
+13-point lift for gemini-2.5-flash on code-review-bench
Test summary
This benchmark evaluates gemini-2.5-flash against grok-3 on the code-review-bench v2 dataset (5 rows), scored by an LLM-as-judge (claude-haiku-4-5). gemini-2.5-flash reached a 47% pass rate versus 34% for grok-3 — a +13-point lift.
- Base model
- grok-3
- via xai · 34% pass rate
- Refined model
- gemini-2.5-flash
- via google · 47% pass rate · 2.8s p95
Base pass rate
34%
Refined pass rate
47%
Refined p95
2.8s
Est. tokens saved
4.1K
+0pts
refined beats base on 5 rows
0%
Base model0%
Invoked-refined▲2 improved▼0 regressed●3 unchanged
Per-row results
1Flag an off-by-one in a pagination loop0% → 100%+100
2Detect a missing await on an async DB write100% → 100%0
3Note a broken null-check after a refactor0% → 0%0
4Find a resource leak on an early-return path0% → 0%0
5Spot a race in a read-modify-write on shared state0% → 100%+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