SAD - Facts (Situating Prompt): leaderboard
SAD factual self-knowledge tasks ask about language models, model identity and differences between human and model properties. 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 Mini (2024-09-12) | 76.19 | #412 |
| 2 | GPT-4 Preview (0125) | 72.61 | #500 |
| 3 | O1 Preview (2024-09-12) | 72.49 | #214 |
| 4 | GPT-4o | 71.84 | #357 |
| 5 | GPT-4 (0613) | 69.62 | #493 |
| 6 | Claude 3.5 Sonnet | 67.53 | #362 |
| 7 | Claude 3 Sonnet | 67.13 | #774 |
| 8 | Claude 3 Opus | 66.54 | #565 |
| 9 | Llama 3 Chat 70B | 64.11 | #772 |
| 10 | Claude 3 Haiku | 62.68 | #801 |
| 11 | Claude 2.1 | 61.97 | #833 |
| 12 | Claude Instant 1.2 | 58.89 | #1003 |
| 13 | Llama 2 70B Chat | 58.34 | #1275 |
| 14 | GPT-3.5 Turbo (0613) | 57.26 | #1106 |
| 15 | Llama 2 13B Chat | 54.11 | #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-facts-situating-prompt · How It Works · Data refreshed daily, snapshot 2026-11-05.