SAD - Influence (Situating Prompt): leaderboard
SAD questions about whether the model can affect events through its outputs, testing understanding of its causal influence. 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) | 93.16 | #214 |
| 2 | O1 Mini (2024-09-12) | 88.09 | #412 |
| 3 | GPT-4o | 87.19 | #357 |
| 4 | GPT-4 Preview (0125) | 86.56 | #500 |
| 5 | GPT-4 (0613) | 85.31 | #493 |
| 6 | Claude 3.5 Sonnet | 82.97 | #362 |
| 7 | Claude 3 Opus | 80.78 | #565 |
| 8 | Llama 3 Chat 70B | 77.19 | #772 |
| 9 | Claude Instant 1.2 | 76.25 | #1003 |
| 10 | Claude 3 Sonnet | 74.38 | #774 |
| 11 | Claude 3 Haiku | 72.81 | #801 |
| 12 | Llama 2 70B | 71.25 | #1039 |
| 13 | Claude 2.1 | 70.94 | #833 |
| 14 | Llama 2 70B Chat | 69.06 | #1275 |
| 15 | Llama 2 13B Chat | 60 | #1414 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sad-influence-situating-prompt · How It Works · Data refreshed daily, snapshot 2026-11-05.