SAD - Influence: leaderboard
SAD questions about whether the model can affect events through its outputs, testing understanding of its causal influence. Uses the plain prompt condition.
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) | 84.38 | #214 |
| 2 | O1 Mini (2024-09-12) | 79.69 | #412 |
| 3 | Claude 3.5 Sonnet | 76.41 | #362 |
| 4 | Claude 3 Opus | 74.69 | #565 |
| 5 | GPT-4o | 67.81 | #357 |
| 6 | Llama 3 Chat 70B | 65.31 | #772 |
| 7 | Claude 3 Sonnet | 64.84 | #774 |
| 8 | Claude Instant 1.2 | 64.84 | #1003 |
| 9 | GPT-4 (0613) | 63.59 | #493 |
| 10 | Claude 2.1 | 63.13 | #833 |
| 11 | GPT-4 Preview (0125) | 62.03 | #500 |
| 12 | Claude 3 Haiku | 61.09 | #801 |
| 13 | Llama 2 70B | 56.25 | #1039 |
| 14 | Llama 2 70B Chat | 55.63 | #1275 |
| 15 | Llama 2 7B | 53.13 | #1498 |
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 · How It Works · Data refreshed daily, snapshot 2026-11-05.