OpenAI GPT-6 Sol & Luna Launch - Warning Circumvention: leaderboard
Alignment chart from OpenAI's GPT-6 Sol and GPT-6 Luna launch post (2026-09-22): share of professional-work rollouts, starting after the environment raises a restriction such as an access-denied error or content policy block, in which the model looks for another way around it. Percent, lower is better; maximum reasoning effort, without system-level controls. Mostly low-stakes situations.
Metric: Circumvention attempt rate (%). Source: openai.com. Source: openai.com. Saturation forecast: Around April 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 17.42 | #5 (GPT-6 Astra) |
| 2 | GPT-6.1 Sol (Max) | 23.48 | #6 (GPT-6.1 Sol) |
| 3 | GPT-6 Luna (Max) | 42.37 | #62 (GPT-6 Luna) |
| 4 | GPT-6 Sol (Max) | 64.39 | #17 (GPT-6 Sol) |
| 5 | GPT-5.6 Sol (Max) | 68.18 | #22 (GPT-5.6 Sol) |
| 6 | GPT-5.6 Luna (Max) | 76.52 | #80 (GPT-5.6 Luna) |
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-sol-luna-launch-warning-circumvention · How It Works · Data refreshed daily, snapshot 2026-11-07.