SAD - Facts: leaderboard
SAD factual self-knowledge tasks ask about language models, model identity and differences between human and model properties. 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 Mini (2024-09-12) | 77.48 | #412 |
| 2 | O1 Preview (2024-09-12) | 73.9 | #214 |
| 3 | GPT-4o | 65.77 | #357 |
| 4 | Claude 3.5 Sonnet | 63.89 | #362 |
| 5 | Claude 3 Opus | 62.76 | #565 |
| 6 | Claude 3 Sonnet | 61.47 | #774 |
| 7 | Claude 2.1 | 60.08 | #833 |
| 8 | GPT-4 Preview (0125) | 58.91 | #500 |
| 9 | GPT-4 (0613) | 56.93 | #493 |
| 10 | Llama 3 Chat 70B | 56.57 | #772 |
| 11 | Claude Instant 1.2 | 53.15 | #1003 |
| 12 | Claude 3 Haiku | 52.77 | #801 |
| 13 | GPT-3.5 Turbo (0613) | 51.09 | #1106 |
| 14 | Llama 2 70B Chat | 50.54 | #1275 |
| 15 | Llama 2 13B Chat | 48.63 | #1414 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sad-facts · How It Works · Data refreshed daily, snapshot 2026-11-05.