Chatbot Arena ELO measures relative user preference from blind head-to-head votes in LMArena. It is a live human-judgment signal for conversational quality under real prompts.
Large-scale blinded pairwise voting reduces single-rater bias; continuous updates surface quality shifts quickly after model releases.
Voter population is self-selected and may not match enterprise or domain users; style, verbosity, and safety tone can influence votes independently of factual correctness. Data: Arena (lmarena.ai) leaderboard dataset, https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset, licensed CC-BY-4.0; rows are converted to the ELO fields shown here.
Higher is better. Scores are ELO-like ratings (typically 900-1400). Compare relative ranking — absolute values shift as new models enter.
Voter population is self-selected and may not match enterprise or domain users.
| # | Model | Score |
|---|---|---|
| 1 | 4203.0 | |
| 2 | 3412.0 | |
| 3 | 2391.0 | |
| 4 | 1539.0 | |
| 5 | 1528.0 | |
| 6 | 1513.0 | |
| 7 | 1512.0 | |
| 8 | 1506.0 | |
| 9 | 1504.0 | |
| 10 | 1503.0 | |
| 11 | 1502.0 | |
| 12 | 1502.0 | |
| 13 | 1501.0 | |
| 14 | 1500.0 | |
| 15 | 1496.0 | |
| 16 | 1494.0 | |
| 17 | 1494.0 | |
| 18 | 1493.0 | |
| 19 | 1492.0 | |
| 20 | 1490.0 | |
| 21 | 1489.0 | |
| 22 | 1485.0 | |
| 23 | 1485.0 | |
| 24 | 1484.0 | |
| 25 | 1483.0 | |
| 26 | 1482.0 | |
| 27 | 1482.0 | |
| 28 | 1481.0 | |
| 29 | 1481.0 | |
| 30 | 1480.0 | |
| 31 | 1480.0 | |
| 32 | 1480.0 | |
| 33 | 1479.0 | |
| 34 | 1478.0 | |
| 35 | 1475.0 | |
| 36 | 1475.0 | |
| 37 | 1475.0 | |
| 38 | 1474.0 | |
| 39 | 1471.0 | |
| 40 | 1471.0 | |
| 41 | 1470.0 | |
| 42 | 1470.0 | |
| 43 | 1469.0 | |
| 44 | 1469.0 | |
| 45 | 1466.0 | |
| 46 | 1466.0 | |
| 47 | 1464.0 | |
| 48 | 1463.0 | |
| 49 | 1463.0 | |
| 50 | 1462.0 |