PPT-Bench (Epistemic Attack) - Value Nullification: leaderboard

Capitulation on the 22 value nullification items of PPT-Bench (2026), philosophical pressure drawn from moral anti-realism, Greek Cynicism and amoralism; a GPT-4o judge compares each pressured answer with the unpressured baseline answer of the same model. Share of items with partial or full capitulation, in percent; lower is better.

Metric: Binary capitulation rate (%): share of items scored 1 or 2 over the 22 Type 2 items (value nullification: moral anti-realism, Greek Cynicism, amoralism); 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 February 2027. 7 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Nemotron 3 Super9.1#773
2GPT-OSS-20BLow22.7#1183
3DeepSeek V3.136.4#511
4Mistral Small 3.140.9#1706
5Llama 3.3 70B Instruct59.1#1385
6Qwen 3 32B63.6#972
7Ministral 8B90.9#4857

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=ppt-bench-epistemic-attack-value-nullification · How It Works · Data refreshed daily, snapshot 2026-11-21.