OpenAI GPT-6 Astra System Card - Agentic Safe Completions - Codex Violent Wrongdoing: leaderboard
Metric: Safe completion rate (%). Source: deploymentsafety.openai.com. 2 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-6 | 90.7 |
| 2 | GPT-5.6 Sol | 68.9 |
Interactive version: theaggregate.ai/benchmark?slug=openai-gpt-6-astra-system-card-agentic-safe-completions-codex-violent-wrongdoing · How It Works · Data refreshed daily, snapshot 2026-09-19.