LoFa - NQ2: leaderboard
Metric: Logical Fallacy Resistance after three attack rounds (LFR@3, %): among the questions a model first answers correctly and whose plain false claim it rejects, the share it still answers correctly after up to three rounds of a DeepSeek-R1-written argument using one of ten logical fallacies, mean over the ten fallacy types, three runs; higher is better; 489 Natural Questions multiple-choice items (second subset). Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V3 | 93.97 |
| 2 | GPT-4 | 90.75 |
| 3 | Llama 3.1 405B | 83.33 |
| 4 | Llama 3.1 70B | 80.08 |
| 5 | GPT-3.5 Turbo | 73.44 |
| 6 | Llama 3.1 8B | 73.38 |
Interactive version: theaggregate.ai/benchmark?slug=lofa-nq2 · How It Works · Data refreshed daily, snapshot 2026-09-29.