Trapped in Your Story: leaderboard
The model picks between two options on 500 debatable questions, then in fresh conversations a user argues for the other option with increasingly personal narratives; the score is the share of questions where it switches at any level. It varies the justification while holding the user's stance fixed, unlike bare-opinion sycophancy tests.
Metric: Narrative-induced stance reversal at any level (%); lower is better. Source: arxiv.org. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 18.75 | #27 |
| 2 | Gemini 3.8 Flash | 21.8 | #36 |
| 3 | GPT-5.5 | 55.6 | #49 |
| 4 | GPT-5.1 | 69.14 | #229 |
| 5 | DeepSeek V4 Pro (Non-reasoning) | 99.8 | #188 (DeepSeek V4 Pro) |
Interactive version: theaggregate.ai/benchmark?slug=trapped-in-your-story · How It Works · Data refreshed daily, snapshot 2026-11-13.