Constitution Adherence Audit - OpenAI Model Spec: leaderboard
The same Petri audit method applied to OpenAI's Model Spec, with Claude Sonnet 4.5 as auditor and OpenAI models under several reasoning settings; the score is the share of transcripts with a confirmed spec violation.
Metric: Share of audited transcripts with a confirmed specification violation (%). Source: arxiv.org. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 2.5 | #176 |
| 2 | GPT-5.2 (Medium) | 3.6 | #176 (GPT-5.2) |
| 3 | GPT-5.1 (Medium) | 3.9 | #229 (GPT-5.1) |
| 4 | GPT-5 (Medium) | 5.1 | #167 (GPT-5) |
| 5 | Claude Sonnet 4.6 | 5.6 | #161 |
| 6 | Gemini 3 Pro | 6.1 | #137 |
| 7 | GPT-5.2 (Low) | 7.1 | #176 (GPT-5.2) |
| 8 | GPT-4o | 11.7 | #741 |
Interactive version: theaggregate.ai/benchmark?slug=constitution-adherence-audit-openai-model-spec · How It Works · Data refreshed daily, snapshot 2026-11-13.