Prove the lift · AI model performance evaluation

llama-3.1-405b vs llama-4-maverick

+11-point lift for llama-3.1-405b on code-review-bench

Test summary

This benchmark evaluates llama-3.1-405b against llama-4-maverick on the code-review-bench v2 dataset (5 rows), scored by an LLM-as-judge (gpt-4o-mini). llama-3.1-405b reached a 56% pass rate versus 45% for llama-4-maverick — a +11-point lift.

Base model
llama-4-maverick
via meta · 45% pass rate
Refined model
llama-3.1-405b
via meta · 56% pass rate · 1.0s p95
Base pass rate
45%
Refined pass rate
56%
Refined p95
1.0s
Est. tokens saved
7.3K
+0pts
refined beats base on 5 rows
0%
Base model
0%
Invoked-refined
0 improved1 regressed4 unchanged

Per-row results

1Note a broken null-check after a refactor100%100%0
2Find a resource leak on an early-return path100%0%-100
3Catch an unsanitized input reaching a SQL string0%0%0
4Spot a race in a read-modify-write on shared state100%100%0
5Identify an N+1 query in an ORM call site100%100%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