Prove the lift · AI model performance evaluation
qwen2.5-vl-32b vs nova-pro
+20-point lift for qwen2.5-vl-32b on code-review-bench
Test summary
This benchmark evaluates qwen2.5-vl-32b against nova-pro on the code-review-bench v1 dataset (7 rows), scored by an LLM-as-judge (claude-haiku-4-5). qwen2.5-vl-32b reached a 62% pass rate versus 42% for nova-pro — a +20-point lift.
- Base model
- nova-pro
- via aws · 42% pass rate
- Refined model
- qwen2.5-vl-32b
- via qwen · 62% pass rate · 3.0s p95
Base pass rate
42%
Refined pass rate
62%
Refined p95
3.0s
Est. tokens saved
12.5K
+0pts
refined beats base on 7 rows
0%
Base model0%
Invoked-refined▲2 improved▼1 regressed●4 unchanged
Per-row results
1Note a broken null-check after a refactor100% → 100%0
2Spot a race in a read-modify-write on shared state100% → 100%0
3Find a resource leak on an early-return path0% → 0%0
4Identify an N+1 query in an ORM call site0% → 100%+100
5Catch an unsanitized input reaching a SQL string100% → 100%0
6Detect a missing await on an async DB write0% → 100%+100
7Flag an off-by-one in a pagination loop100% → 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