Prove the lift · AI model performance evaluation
llama-3.3-70b vs gemma-3-27b
+10-point lift for llama-3.3-70b on code-review-bench
Test summary
This benchmark evaluates llama-3.3-70b against gemma-3-27b on the code-review-bench v2 dataset (7 rows), scored by an LLM-as-judge (gpt-4o-mini). llama-3.3-70b reached a 39% pass rate versus 29% for gemma-3-27b — a +10-point lift.
- Base model
- gemma-3-27b
- via google · 29% pass rate
- Refined model
- llama-3.3-70b
- via meta · 39% pass rate · 2.2s p95
Base pass rate
29%
Refined pass rate
39%
Refined p95
2.2s
Est. tokens saved
3.2K
+0pts
refined beats base on 7 rows
0%
Base model0%
Invoked-refined▲3 improved▼0 regressed●4 unchanged
Per-row results
1Flag an off-by-one in a pagination loop0% → 100%+100
2Detect a missing await on an async DB write0% → 0%0
3Find a resource leak on an early-return path0% → 0%0
4Note a broken null-check after a refactor0% → 0%0
5Identify an N+1 query in an ORM call site100% → 100%0
6Spot a race in a read-modify-write on shared state0% → 100%+100
7Catch an unsanitized input reaching a SQL string0% → 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