LLM Position Bias First-Shown Pick Rate
Chat & Writing · 2026-04-21
First-Shown Pick Rate is the share of decisive prompt responses where the model chose the version displayed first. The unbiased target is 50%; values above or below 50% indicate first- or second-position preference.
Top models (closest to 50.0 is better)
| Model | Score |
|---|---|
| Trinity Large Thinking | 48.9 |
| Doubao Seed 2.0 Pro | 48.1 |
| Gemini 3.1 Flash-Lite Preview | 52.5 |
| DeepSeek-V3.2 | 53.8 |
| MiMo-V2-Pro | 54.5 |
| Qwen3.5 122B A10B | 56.7 |
| MiniMax M2.7 | 58.2 |
| Claude Fable 5 | 58.9 |
| GLM-5.1 | 59.8 |
| Opus 4.8 | 60.7 |
| Gemini 3.5 Flash | 62.1 |
| Qwen3.6 Plus (2026-04-02) | 64.3 |
| Opus 4.6 | 64.4 |
| Sonnet 4.6 | 65.1 |
| Qwen3.7-Max | 65.3 |
| MiniMax M3 | 65.3 |
| Opus 4.7 | 65.6 |
| Grok 4.20 | 65.8 |
| Qwen3.5 397B A17B | 65.8 |
| Gemini 3.1 Pro Preview | 66.0 |
| ERNIE 5.0 | 66.1 |
| DeepSeek-V4-Pro | 66.7 |
| Mistral Medium 3.1 | 67.2 |
| GPT-5.5 | 69.9 |
| Kimi K2.6 | 72.5 |
| Mistral Large 3 675B Instruct 2512 | 27.4 |
| GPT-5.4 | 72.7 |
| Gemma 4 31B IT | 74.6 |
| GPT-5.4 Mini | 75.7 |
| Kimi K2.5 | 75.8 |
| Mistral Medium 3.5 | 82.8 |