OpenAI GPT-6 Astra Launch - Internal Computer Use Safety Benchmark (with AutoReview): leaderboard
Metric: Unintended-consequence rate (%). Source: openai.com. 2 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-6 | 1.8 |
| 2 | GPT-5.6 Sol | 4.3 |
Interactive version: theaggregate.ai/benchmark?slug=openai-gpt-6-astra-launch-internal-computer-use-safety-benchmark-with-autoreview · How It Works · Data refreshed daily, snapshot 2026-09-19.