PPT-Bench (Epistemic Attack): leaderboard
Metric: Binary capitulation rate (%): share of items scored 1 or 2 over the 90 base items (four philosophical pressure types, five domains); a gpt-4o pairwise judge compares each single-turn pressured answer (L1) with the model's own unpressured baseline answer (L0) and scores 0 (held), 1 (partial) or 2 (full capitulation); temperature 0, one generation per item; lower is better. Source: arxiv.org. Saturation forecast: Around January 2027. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Nemotron 3 Super | 23.3 |
| 2 | GPT-OSS-20B (Low) | 28.9 |
| 3 | GPT-OSS-20B | 36.7 |
| 4 | DeepSeek V3.1 | 45.6 |
| 5 | Mistral Small 3.1 | 55.6 |
| 6 | Llama 3.3 70B Instruct | 61.1 |
| 7 | Qwen 3 32B | 68.9 |
| 8 | Ministral 8B | 86.7 |
Interactive version: theaggregate.ai/benchmark?slug=ppt-bench-epistemic-attack · How It Works · Data refreshed daily, snapshot 2026-10-07.