LLM Sycophancy Stripped Rate
Safety · 2026-03-08
Stripped sycophancy rate measures sycophancy after removing user-opinion cues from prompts. Lower values indicate less sycophancy.
Top models (lower is better)
| Model | Score |
|---|---|
| Gemini 3.1 Pro Preview | 0.0 |
| Grok 4.3 | 0.0 |
| Gemini 3.5 Flash | 0.0 |
| Gemini 3.1 Flash-Lite Preview | 0.5 |
| Claude Fable 5 | 2.0 |
| Grok 4.20 | 2.0 |
| Opus 4.7 | 2.0 |
| MiMo-V2.5-Pro | 2.5 |
| Kimi K2.6 | 3.0 |
| Kimi K2.5 | 3.0 |
| GPT-5.5 | 3.5 |
| GLM-5.1 | 3.5 |
| Qwen3.5 397B A17B | 4.0 |
| DeepSeek-V4-Pro | 4.0 |
| DeepSeek-V3.2 | 4.0 |
| Qwen3.6-Max-Preview | 4.5 |
| Gemma 4 31B IT | 5.0 |
| Qwen3.7-Plus | 5.0 |
| Opus 4.6 | 6.0 |
| MiniMax M2.5 | 6.0 |
| MiniMax M3 | 7.0 |
| ERNIE 5.1 | 7.5 |
| Sonnet 4.6 | 7.5 |
| ERNIE 5.0 | 8.0 |
| Hy3 preview | 10.1 |
| GLM-5 | 10.1 |
| MiniMax M2.7 | 14.1 |
| Trinity Large Thinking | 15.1 |
| Doubao Seed 2.0 Pro | 17.1 |
| GPT-4.1 | 18.1 |
| Mistral Medium 3.5 | 24.1 |
| Mistral Large 3 675B Instruct 2512 | 31.2 |