OpenAI GPT-5.6 System Card - Challenging Prompts - Self-Harm: leaderboard
Table 1 (production benchmarks with challenging prompts) of OpenAI's GPT-5.6 System Card (Deployment Safety Hub, 2026-07-09): not_unsafe rate on deliberately difficult self-harm prompts drawn from production-like traffic. Percent (the card prints 0-1), higher is better. Seven GPT models from GPT-5.1 Thinking to GPT-5.6 Luna.
Metric: not_unsafe rate (%). Source: deploymentsafety.openai.com. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (Thinking) | 98.7 | #83 (GPT-5.4) |
| 2 | GPT-5.6 Terra | 96.2 | #40 |
| 3 | GPT-5.2 (Thinking) | 96.1 | #115 (GPT-5.2) |
| 4 | GPT-5.6 Luna | 95.4 | #79 |
| 5 | GPT-5.6 Sol | 94.5 | #22 |
| 6 | GPT-5.1 (Thinking) | 92.6 | #145 (GPT-5.1) |
| 7 | GPT-5.5 | 91.7 | #33 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openai-gpt-5-6-system-card-challenging-prompts-self-harm · How It Works · Data refreshed daily, snapshot 2026-11-05.