PPT-Bench (Epistemic Attack) - Full Capitulation: leaderboard
Metric: Full capitulation rate (%): share of items scored 2 over the 90 base items; 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: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-OSS-20B | 1.1 |
| 2 | DeepSeek V3.1 | 2.2 |
| 3 | GPT-OSS-20B (Low) | 2.2 |
| 4 | Mistral Small 3.1 | 5.6 |
| 5 | Llama 3.3 70B Instruct | 6.7 |
| 6 | Qwen 3 32B | 6.7 |
| 7 | Nemotron 3 Super | 10 |
| 8 | Ministral 8B | 12.2 |
Interactive version: theaggregate.ai/benchmark?slug=ppt-bench-epistemic-attack-full-capitulation · How It Works · Data refreshed daily, snapshot 2026-10-07.