Situational Awareness Dataset (SAD): leaderboard
Behavioral benchmark for LLM situational awareness, testing self-knowledge, self-recognition, stage-of-deployment inference, and actions conditioned on model self-knowledge across 7 categories and 16 tasks.
Metric: Score (%, higher is better). Source: situational-awareness-dataset.org. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 Preview (2024-09-12) | 59.92 | #197 |
| 2 | Claude 3.5 Sonnet | 53.86 | #350 |
| 3 | O1 Mini (2024-09-12) | 53.3 | #403 |
| 4 | Claude 3 Opus | 50.27 | #571 |
| 5 | Claude 3 Sonnet | 47.19 | #775 |
| 6 | GPT-4o | 46.37 | #343 |
| 7 | Claude 2.1 | 44.3 | #839 |
| 8 | GPT-4 Preview (0125) | 43.2 | #486 |
| 9 | Claude Instant 1.2 | 42.72 | #1017 |
| 10 | GPT-4 (0613) | 42.18 | #491 |
| 11 | Claude 3 Haiku | 41.38 | #803 |
| 12 | Llama 2 70B Chat | 36.58 | #1237 |
| 13 | GPT-3.5 Turbo (0613) | 35.83 | #1118 |
| 14 | Llama 2 13B Chat | 34.9 | #1465 |
| 15 | Llama 2 7B | 32.88 | #1520 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=situational-awareness-dataset-sad · How It Works · Data refreshed daily, snapshot 2026-10-20.