Beyond the Sycophancy Score - Contested Policy: leaderboard
The model is asked a contested policy or moral question, then the user pushes one side for three turns; the number is the share of eligible conversations (those whose first reply had not already taken that position) where the final reply adopts the user's side. It measures giving up neutrality under social pressure rather than abandoning a fact.
Metric: Final-turn adoption of the user's side on a contested policy or moral question (%). Source: arxiv.org. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.3 (Non-reasoning) | 0.4 | #333 (Grok 4.3) |
| 2 | Claude Sonnet 5 (Non-reasoning) | 2.3 | #111 (Claude Sonnet 5) |
| 3 | GLM-5.2 (Non-reasoning) | 6.6 | #138 (GLM-5.2) |
| 4 | GPT-5.6 Luna (Non-reasoning) | 16.8 | #121 (GPT-5.6 Luna) |
| 5 | Qwen 3.5 397B A17B (Non-reasoning) | 21.4 | #246 (Qwen 3.5 397B A17B) |
| 6 | DeepSeek V4 Flash (0731) (Non-reasoning) | 51.4 | #139 (DeepSeek V4 Flash (0731)) |
Interactive version: theaggregate.ai/benchmark?slug=beyond-the-sycophancy-score-contested-policy · How It Works · Data refreshed daily, snapshot 2026-11-13.