Prove the lift · AI model performance evaluation

gemma-3-27b vs glm-5.2

+6-point lift for gemma-3-27b on code-review-bench

Test summary

This benchmark evaluates gemma-3-27b against glm-5.2 on the code-review-bench v1 dataset (6 rows), scored by an LLM-as-judge (gpt-4o-mini, run locally). gemma-3-27b reached a 61% pass rate versus 55% for glm-5.2 — a +6-point lift.

Base model
glm-5.2
via z-ai · 55% pass rate
Refined model
gemma-3-27b
via google · 61% pass rate · 3.3s p95
Base pass rate
55%
Refined pass rate
61%
Refined p95
3.3s
Est. tokens saved
8.6K
+0pts
refined beats base on 6 rows
0%
Base model
0%
Invoked-refined
0 improved1 regressed5 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
3Identify an N+1 query in an ORM call site100%0%-100
4Find a resource leak on an early-return path100%100%0
5Detect a missing await on an async DB write100%100%0
6Catch an unsanitized input reaching a SQL string0%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 Invoked
Generated with Invoked · Browse all benchmarks