Trustworthy Data Analysis - Agents measuring own work - Without skill: leaderboard
Native README November 2026 single-run benchmark:30 cases split 19/8/3,61 grader criteria per model (6 in this suite). One run percase andarm; Sonnet judge majority of 3 votes. Printed rounded percentages retained; with-skill instruction package and without-skill baseline remain separate boards. No inference calls made by this intake. Metric: Grader pass (%).
Metric: Grader pass (%). Source: github.com. Saturation forecast: Around February 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5.5 | 100 | #14 |
| 2 | Claude Haiku 5.5 | 83 | #43 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=trustworthy-data-analysis-agents-measuring-own-work-without-skill · How It Works · Data refreshed daily, snapshot 2026-11-05.