SycoEval-EM - Acquiescence: leaderboard
Acquiescence to inappropriate patient requests across 75 simulated clinical encounters per model. October 2025 provider routes, paper v4 Table 2, lower is better. This is a specific simulated behavior, not overall clinical competence; reported uncertainty is not another score. This series reports Acquiescence Rate (%). Source: https://arxiv.org/html/2601.16529v4.
Metric: Acquiescence Rate (%). Source: arxiv.org. Saturation forecast: Not forecast. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 0 | #148 |
| 2 | Grok 3 Mini | 0 | #289 |
| 3 | Grok 4 | 1.3 | #189 |
| 4 | Claude 3.5 Haiku | 1.3 | #571 |
| 5 | Grok 4 Fast | 1.3 | #266 |
| 6 | Kimi K2 | 8 | #252 |
| 7 | GPT-4.1 Nano | 9.3 | #730 |
| 8 | GPT-5 | 17.3 | #105 |
| 9 | GPT-5 Nano | 30.7 | #439 |
| 10 | Gemini 2.5 Flash | 36 | #255 |
| 11 | Gemini 2.5 Pro | 46.7 | #159 |
| 12 | GPT-4o Mini | 46.7 | #598 |
| 13 | GPT-5 Mini | 46.7 | #193 |
| 14 | GLM 4.5 Air | 52 | #403 |
| 15 | DeepSeek V3.1 | 53.3 | #275 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sycoeval-em-acquiescence · How It Works · Data refreshed daily, snapshot 2026-11-05.