SAD - Stages: leaderboard
SAD deployment-stage inference asks models to identify pretraining, supervised training, evaluation and deployment contexts from prompt evidence. 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) | 55.02 | #214 |
| 2 | Claude 3 Opus | 50.69 | #565 |
| 3 | O1 Mini (2024-09-12) | 50.48 | #412 |
| 4 | GPT-4 (0613) | 50.17 | #493 |
| 5 | Llama 3 Chat 70B | 46.91 | #772 |
| 6 | Claude 3.5 Sonnet | 46.47 | #362 |
| 7 | Claude 3 Sonnet | 46.31 | #774 |
| 8 | GPT-4 Preview (0125) | 45 | #500 |
| 9 | Claude Instant 1.2 | 41 | #1003 |
| 10 | Claude 2.1 | 40.81 | #833 |
| 11 | GPT-3.5 Turbo (0613) | 39.88 | #1106 |
| 12 | GPT-4o | 39.75 | #357 |
| 13 | Claude 3 Haiku | 39.64 | #801 |
| 14 | Llama 2 13B Chat | 39.64 | #1414 |
| 15 | Llama 2 70B Chat | 38.44 | #1275 |
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 · How It Works · Data refreshed daily, snapshot 2026-11-05.