Prove the lift · AI model performance evaluation

gemini-3-pro vs deepseek-r1

+6-point lift for gemini-3-pro on code-review-bench

Test summary

This benchmark evaluates gemini-3-pro against deepseek-r1 on the code-review-bench v1 dataset (7 rows), scored by an LLM-as-judge (gpt-4o-mini, run locally). gemini-3-pro reached a 57% pass rate versus 51% for deepseek-r1 — a +6-point lift.

Base model
deepseek-r1
via deepseek · 51% pass rate
Refined model
gemini-3-pro
via google · 57% pass rate · 3.2s p95
Base pass rate
51%
Refined pass rate
57%
Refined p95
3.2s
Est. tokens saved
5.3K
+0pts
refined beats base on 7 rows
0%
Base model
0%
Invoked-refined
▲2 improved▼0 regressed●5 unchanged

Per-row results

1Spot a race in a read-modify-write on shared state0% → 0%0
2Find a resource leak on an early-return path0% → 100%+100
3Identify an N+1 query in an ORM call site0% → 0%0
4Detect a missing await on an async DB write0% → 0%0
5Catch an unsanitized input reaching a SQL string100% → 100%0
6Flag an off-by-one in a pagination loop100% → 100%0
7Note a broken null-check after a refactor0% → 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 Invoked
Generated with Invoked · Browse all benchmarks