FeatureLoop: leaderboard

Coding agents implement a feature and respond as its requirements evolve. Executable checks measure completed tasks, partial credit, repairs and regressions across the same persistent workspace.

Source: arxiv.org.

Benchmark results

Results are published separately for each benchmark or evaluation setting.

Interactive version: theaggregate.ai/benchmark?slug=featureloop · How It Works · Data refreshed daily, snapshot 2026-11-17.