OpenAI GPT-6 Astra System Card - Static Jailbreaks - Bio Severe: leaderboard
Table 5 (static jailbreak evaluations) of OpenAI's GPT-6 Astra System Card (Deployment Safety Hub, 2026-09-03): share of fixed adversarial severe biology jailbreak prompts the model refuses, without production safeguards. Percent (the card prints 0-1), higher is better. Attacker models generated the prompts against earlier GPT models, so GPT-5.5 Thinking and GPT-5.6 Sol scores are depressed by construction. GPT-6 Sol and GPT-6 Luna come from Table 27 of the card's section 11 (GPT-6 Sol and GPT-6 Luna, added 2026-09-22), which prints percentages and re-prints the other cells.
Metric: Jailbreak refusal rate (%). Source: deploymentsafety.openai.com. Source: deploymentsafety.openai.com. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 98.2 | #6 |
| 2 | GPT-6.1 Sol | 94.3 | #7 |
| 3 | GPT-6 Sol | 81.3 | #19 |
| 4 | GPT-6 Luna | 73 | #61 |
| 5 | GPT-5.5 | 12.1 | #33 |
| 6 | GPT-5.6 Sol | 10.3 | #22 |
No result here: #3 Claude Opus 5.5, #9 Gemini 4 Argon, #10 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=openai-gpt-6-astra-system-card-static-jailbreaks-bio-severe · How It Works · Data refreshed daily, snapshot 2026-11-05.