Sycophancy detection on SYCON-Bench raw n=100
68Sycophancy RateGemini 3 Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini 3 FlashFamily=Gemini, Organization=Google2026.04 | 68 | 0.97 | |
| Llama 4 Maverick†Family=Llama, Organization=Meta, Outlier=Discussed in text2026.04 | 58 | 0.76 | |
| GPT-5.2Family=GPT, Organization=OpenAI2026.04 | 53 | 0.72 | |
| Gemini 3 ProFamily=Gemini, Organization=Google2026.04 | 53 | 0.99 | |
| GPT-4.1Family=GPT, Organization=OpenAI2026.04 | 48 | 0.83 | |
| DeepSeek V3Organization=DeepSeek2026.04 | 47 | 0.74 | |
| GPT-4oFamily=GPT, Organization=OpenAI2026.04 | 34 | 0.65 | |
| Llama 4 ScoutFamily=Llama, Organization=Meta2026.04 | 16 | 0.9 | |
| Sonnet 4.5Family=Claude, Organization=Anthropic2026.04 | 8 | 0.88 | |
| Opus 4.6Family=Claude, Organization=Anthropic2026.04 | 2 | 0.97 |