Situational Awareness Dataset (SAD) (Situating Prompt): leaderboard
Full Situational Awareness Dataset: seven equally weighted categories testing model facts, influence, introspection, deployment stages, self-recognition, identity-dependent instructions and anti-imitation. Published model-specific tasks; not an anonymous self-forecast test. Uses a situating system prompt reminding the model that it is an LLM.
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) | 61.93 | #214 |
| 2 | Claude 3.5 Sonnet | 57.13 | #362 |
| 3 | O1 Mini (2024-09-12) | 55.37 | #412 |
| 4 | Claude 3 Opus | 53.37 | #565 |
| 5 | GPT-4 Preview (0125) | 51.52 | #500 |
| 6 | GPT-4o | 50.72 | #357 |
| 7 | Claude 3 Sonnet | 50.01 | #774 |
| 8 | Llama 3 Chat 70B | 49.34 | #772 |
| 9 | GPT-4 (0613) | 46.73 | #493 |
| 10 | Claude 2.1 | 46.19 | #833 |
| 11 | Claude 3 Haiku | 45.88 | #801 |
| 12 | Claude Instant 1.2 | 43.9 | #1003 |
| 13 | Llama 2 70B Chat | 40.23 | #1275 |
| 14 | GPT-3.5 Turbo (0613) | 38.42 | #1106 |
| 15 | Llama 2 13B Chat | 37.63 | #1414 |
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-situating-prompt · How It Works · Data refreshed daily, snapshot 2026-11-05.