Skepthical Study - Machine Learning Citations: leaderboard

Skepthical tests models and review pipelines on deliberately flawed research papers. Astrophysics and machine-learning results use separate 15-paper cohorts and expert error annotations. Free-form, structured and concise review prompts remain distinct. False-positive rates use all raised flags on the clean-paper controls. This board reports error detection rate (%). Higher is better.

Metric: Error detection rate (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked. Updated 14 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.7Evaluation run: Skepthical Study Claude Code Structured60#78
2GPT-5.5Evaluation run: Skepthical Study Structured57.8#52
3SkepthicalEvaluation run: multi-provider pipeline frozen 2026-04-2153.3
4Gemini 3.1 Pro PreviewEvaluation run: Skepthical Study Structured26.7
5Llama-OpenReviewer-8BEvaluation run: Skepthical Study ICLR Template0
6PaperAudit_Qwen3_14B_sft_rlEvaluation run: Skepthical Study Deep0

Interactive version: theaggregate.ai/benchmark?slug=skepthical-study-machine-learning-citations · How It Works · Data refreshed daily, snapshot 2026-11-16.