SAD - Self-Recognition (Situating Prompt): leaderboard
SAD self-recognition tests distinguishing the model's own generated text from human-written text with similar content. 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 | #214 |
| 2 | O1 Mini (2024-09-12) | 76.13 | #412 |
| 3 | Llama 3 Chat 70B | 73.69 | #772 |
| 4 | GPT-4 Preview (0125) | 70.44 | #500 |
| 5 | GPT-4o | 68.63 | #357 |
| 6 | Claude 3.5 Sonnet | 68.13 | #362 |
| 7 | Claude 3 Opus | 63.28 | #565 |
| 8 | Claude 2.1 | 59.69 | #833 |
| 9 | Llama 2 13B Chat | 56.88 | #1414 |
| 10 | Claude Instant 1.2 | 55.75 | #1003 |
| 11 | GPT-4 (0613) | 54 | #493 |
| 12 | Llama 2 70B Chat | 52.88 | #1275 |
| 13 | Claude 3 Haiku | 51.69 | #801 |
| 14 | GPT-3.5 Turbo (0613) | 50.81 | #1106 |
| 15 | davinci-002 | 50.63 | #1400 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sad-self-recognition-situating-prompt · How It Works · Data refreshed daily, snapshot 2026-11-05.