Prove the lift · AI model performance evaluation
gemini-2.5-pro vs deepseek-r1
+16-point lift for gemini-2.5-pro on code-review-bench
Test summary
This benchmark evaluates gemini-2.5-pro against deepseek-r1 on the code-review-bench v2 dataset (5 rows), scored by an LLM-as-judge (gpt-4o-mini). gemini-2.5-pro reached a 66% pass rate versus 50% for deepseek-r1 — a +16-point lift.
- Base model
- deepseek-r1
- via deepseek · 50% pass rate
- Refined model
- gemini-2.5-pro
- via google · 66% pass rate · 0.9s p95
Base pass rate
50%
Refined pass rate
66%
Refined p95
0.9s
Est. tokens saved
11.2K
+0pts
refined beats base on 5 rows
0%
Base model0%
Invoked-refined▲1 improved▼0 regressed●4 unchanged
Per-row results
1Note a broken null-check after a refactor100% → 100%0
2Flag an off-by-one in a pagination loop100% → 100%0
3Catch an unsanitized input reaching a SQL string0% → 0%0
4Detect a missing await on an async DB write0% → 100%+100
5Identify an N+1 query in an ORM call site0% → 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 InvokedGenerated with Invoked · Browse all benchmarks