RedlineBench — Mercor evaluation (RedlineBench (Mercor))

We captured Mercor's RedlineBench results on October 15, 2026. The table preserves 12 published configurations, their source identifiers, reasoning settings, and disclosed turn-weighted mean score values. These are results from Mercor's evaluation setup; we did not rerun them.

This dated source table is display only and does not enter overall or category rankings. Source configurations retain their external identity, including Pro descriptors and different effort settings.

The score is a turn-weighted mean, not task pass rate. Mercor uses a uniform Claude Code Harbor setup and a three-model judge panel; sample-task grades come from one judge and are not the leaderboard score. The panel replaces the original Gemini 3.1 judge with Gemini 3.5 Flash.

The row labels retain published error margins and sample counts when supplied. Missing margins and counts remain unreported; a sample count is not used to reconstruct a task denominator. Except where the dedicated source defines an interval, the confidence level is unreported. Equal displayed scores and overlapping margins do not establish statistical ties. Capture time does not establish individual evaluation dates.

The general methodology lists expert-authored rubrics, scored criterion by criterion and a judge of 3-LLM judge panel over expert rubrics. Its run-count label is k = 3. The linked result page takes precedence where its setup differs, as noted above.