PreScam - Action Prediction (Limited): leaderboard
Metric: Action HitRate (%): the share of gold future scammer action elements recovered by the prediction, averaged over cases, for a zero-shot forecast of the scammer's remaining actions from the conversation prefix up to a boundary turn chosen per case, Limited setting (the model also receives the exact number of remaining scammer actions); GPT-4o-mini extracts structured labels from the predicted and gold actions; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 5 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.5 | 63.93 |
| 2 | DeepSeek V3.2 | 63.57 |
| 3 | GPT-5 Chat | 57.53 |
| 4 | Llama 3.3 70B Instruct | 52.52 |
| 5 | GPT-4o Mini | 40.45 |
Interactive version: theaggregate.ai/benchmark?slug=prescam-action-prediction-limited · How It Works · Data refreshed daily, snapshot 2026-10-07.