Automated benchmark using 500 challenging user queries from Chatbot Arena. An LLM judge scores model responses against a baseline (GPT-4-0314).
This definition has not completed editorial and provenance review. Verify its metric, version, source, and evaluation configuration upstream before using it for a model-selection decision.
Closely correlated with human preference rankings from Chatbot Arena. Cost-effective automated alternative to crowd evaluation.
LLM judge (typically GPT-4) may have systematic biases. Baseline model affects absolute scores.
Higher is better. Win rate percentage against baseline (0–100%).
LLM judge (typically GPT-4) may have systematic biases.
| # | Model | Score |
|---|---|---|
| 1 | 97.1 | |
| 2 | 95.9 | |
| 3 | 95.8 | |
| 4 | 95.6 | |
| 5 | 95.6 | |
| 6 | 95.2 | |
| 7 | 93.1 | |
| 8 | 92.8 | |
| 9 | 92.7 | |
| 10 | 92.7 | |
| 11 | 92.3 | |
| 12 | 92.3 | |
| 13 | 92.2 | |
| 14 | 92.1 | |
| 15 | 92.0 | |
| 16 | 91.0 | |
| 17 | 91.0 | |
| 18 | 91.0 | |
| 19 | 90.9 | |
| 20 | 90.8 | |
| 21 | 90.5 | |
| 22 | 90.4 | |
| 23 | 90.1 | |
| 24 | 90.0 | |
| 25 | 89.6 | |
| 26 | 89.4 | |
| 27 | 89.4 | |
| 28 | 89.1 | |
| 29 | 89.1 | |
| 30 | 89.1 | |
| 31 | 89.0 | |
| 32 | 88.9 | |
| 33 | 88.8 | |
| 34 | 88.8 | |
| 35 | 88.7 | |
| 36 | 88.6 | |
| 37 | 88.6 | |
| 38 | 88.6 | |
| 39 | 88.5 | |
| 40 | 88.3 | |
| 41 | 88.2 | |
| 42 | 88.1 | |
| 43 | 87.9 | |
| 44 | 87.8 | |
| 45 | 87.5 | |
| 46 | 87.4 | |
| 47 | 87.2 | |
| 48 | 87.2 | |
| 49 | 87.1 | |
| 50 | 87.0 |