Prove the lift · AI model performance evaluation

claude-opus-4-8 vs claude-opus-4-6

+9-point lift for claude-opus-4-8 on code-review-bench

Test summary

This benchmark evaluates claude-opus-4-8 against claude-opus-4-6 on the code-review-bench v1 dataset (6 rows), scored by an LLM-as-judge (gpt-4o-mini, run locally). claude-opus-4-8 reached a 48% pass rate versus 39% for claude-opus-4-6 — a +9-point lift.

Base model
claude-opus-4-6
via anthropic · 39% pass rate
Refined model
claude-opus-4-8
via anthropic · 48% pass rate · 2.8s p95
Base pass rate
39%
Refined pass rate
48%
Refined p95
2.8s
Est. tokens saved
8.3K
+0pts
refined beats base on 6 rows
0%
Base model
0%
Invoked-refined
▲4 improved▼0 regressed●2 unchanged

Per-row results

1Spot a race in a read-modify-write on shared state0% → 100%+100
2Flag an off-by-one in a pagination loop100% → 100%0
3Note a broken null-check after a refactor0% → 100%+100
4Catch an unsanitized input reaching a SQL string0% → 100%+100
5Detect a missing await on an async DB write0% → 0%0
6Identify an N+1 query in an ORM call site0% → 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