Prove the lift · AI model performance evaluation

nemotron-ultra-253b vs gpt-5.1

+22-point lift for nemotron-ultra-253b on code-review-bench

Test summary

This benchmark evaluates nemotron-ultra-253b against gpt-5.1 on the code-review-bench v3 dataset (5 rows), scored by an LLM-as-judge (llama-3.1-8b). nemotron-ultra-253b reached a 66% pass rate versus 44% for gpt-5.1 — a +22-point lift.

Base model
gpt-5.1
via openai · 44% pass rate
Refined model
nemotron-ultra-253b
via nvidia · 66% pass rate · 3.4s p95
Base pass rate
44%
Refined pass rate
66%
Refined p95
3.4s
Est. tokens saved
9.1K
+0pts
refined beats base on 5 rows
0%
Base model
0%
Invoked-refined
▲3 improved▼0 regressed●2 unchanged

Per-row results

1Find a resource leak on an early-return path0% → 100%+100
2Catch an unsanitized input reaching a SQL string100% → 100%0
3Spot a race in a read-modify-write on shared state0% → 0%0
4Identify an N+1 query in an ORM call site0% → 100%+100
5Detect a missing await on an async DB write0% → 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