← All benchmarks

MacroscopeBench Precision

Code · 2026-09-22

MacroscopeBench precision: the share of all findings a reviewer reports across the benchmark that a second LLM judge, which can explore the codebase, rates as valid correctness bugs introduced or exposed by the reviewed commit; style, subjective and out-of-scope findings count against it. It is averaged over three runs, in percent. The denominator is each model's own finding count, so it is a diagnostic rather than a fixed-item score.

Top models (higher is better)