HarnessAudit-Bench: leaderboard
Metric: Overall harness safety score (0-1; task-level safety adherence gate times weighted task completion, action validity and perturbation stability; 210 tasks in eight domains). Source: arxiv.org. Saturation forecast: Rough model projection: around 2027. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 (Claude Code) | 0.32 |
| 2 | Claude Opus 4.6 (Claude Code) | 0.29 |
Interactive version: theaggregate.ai/benchmark?slug=harnessaudit-bench · How It Works · Data refreshed daily, snapshot 2026-09-26.