Prove the lift · AI model performance evaluation
claude-opus-4-6 vs gpt-4-turbo
-3-point lift for claude-opus-4-6 on code-review-bench
Test summary
This benchmark evaluates claude-opus-4-6 against gpt-4-turbo on the code-review-bench v1 dataset (7 rows), scored by an LLM-as-judge (llama-3.1-8b). claude-opus-4-6 reached a 23% pass rate versus 26% for gpt-4-turbo — a -3-point lift.
- Base model
- gpt-4-turbo
- via openai · 26% pass rate
- Refined model
- claude-opus-4-6
- via anthropic · 23% pass rate · 1.9s p95
Base pass rate
26%
Refined pass rate
23%
Refined p95
1.9s
Est. tokens saved
1.2K
−0pts
no lift — the base model held its ground
0%
Base model0%
Invoked-refined▲0 improved▼2 regressed●5 unchanged
Per-row results
1Flag an off-by-one in a pagination loop0% → 0%0
2Note a broken null-check after a refactor0% → 0%0
3Identify an N+1 query in an ORM call site0% → 0%0
4Spot a race in a read-modify-write on shared state100% → 100%0
5Detect a missing await on an async DB write100% → 0%-100
6Catch an unsanitized input reaching a SQL string0% → 0%0
7Find a resource leak on an early-return path100% → 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