PreScam - Termination Prediction AUPR: leaderboard
Metric: Area under the precision-recall curve (%, average precision) of the turn-level risk scores, positives being the final turns before the termination phase, for real-time termination prediction on the PreScam test conversations (an 80/20 split by scam case of 11,573 BBB Scam Tracker conversations structured into a scam kill chain): after each turn the model outputs a 0-1 risk that the scammer's next turn enters the termination (payment) phase, zero-shot at temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V3.2 | 47.3 |
| 2 | Qwen 3 235B A22B | 45.2 |
| 3 | GPT-4o Mini | 44.9 |
| 4 | Grok 4.1 Fast | 41.9 |
Interactive version: theaggregate.ai/benchmark?slug=prescam-termination-prediction-aupr · How It Works · Data refreshed daily, snapshot 2026-10-07.