SAD - Stages (Situating Prompt): leaderboard
SAD deployment-stage inference asks models to identify pretraining, supervised training, evaluation and deployment contexts from prompt evidence. 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) | 61.52 | #214 |
| 2 | O1 Mini (2024-09-12) | 52.73 | #412 |
| 3 | Claude 3 Opus | 51.09 | #565 |
| 4 | Claude 3.5 Sonnet | 49.31 | #362 |
| 5 | Llama 3 Chat 70B | 48.44 | #772 |
| 6 | GPT-4 (0613) | 48.31 | #493 |
| 7 | GPT-4 Preview (0125) | 46.88 | #500 |
| 8 | Claude 3 Sonnet | 44.19 | #774 |
| 9 | GPT-4o | 44.06 | #357 |
| 10 | Llama 2 70B Chat | 42.31 | #1275 |
| 11 | Claude 2.1 | 41.5 | #833 |
| 12 | GPT-3.5 Turbo (0613) | 41.39 | #1106 |
| 13 | Llama 2 13B Chat | 40.83 | #1414 |
| 14 | Claude Instant 1.2 | 39.94 | #1003 |
| 15 | Claude 3 Haiku | 38.95 | #801 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sad-stages-situating-prompt · How It Works · Data refreshed daily, snapshot 2026-11-05.