SAD - Self-Recognition: leaderboard
SAD self-recognition tests distinguishing the model's own generated text from human-written text with similar content. 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) | 88.56 | #214 |
| 2 | O1 Mini (2024-09-12) | 76.47 | #412 |
| 3 | GPT-4o | 72 | #357 |
| 4 | Claude 3.5 Sonnet | 66.25 | #362 |
| 5 | Llama 3 Chat 70B | 65.69 | #772 |
| 6 | Claude 3 Opus | 64.31 | #565 |
| 7 | Claude 2.1 | 62.19 | #833 |
| 8 | Claude Instant 1.2 | 59.81 | #1003 |
| 9 | GPT-4 Preview (0125) | 59.19 | #500 |
| 10 | Llama 2 70B Chat | 54.13 | #1275 |
| 11 | GPT-4 (0613) | 52.09 | #493 |
| 12 | Llama 2 13B Chat | 51.25 | #1414 |
| 13 | GPT-3.5 Turbo (0613) | 51.06 | #1106 |
| 14 | Claude 3 Haiku | 51 | #801 |
| 15 | davinci-002 | 50.56 | #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 · How It Works · Data refreshed daily, snapshot 2026-11-05.