Emotional intelligence benchmark using roleplay scenarios judged by Claude. Measures empathy, social IQ, insight, and analytical depth via Elo ratings from pairwise comparisons.
Unique signal measuring emotional and social intelligence; resistant to gaming; uses pairwise Elo methodology.
Judge model (Claude 3.7 Sonnet) may have biases; Elo depends on comparison pool composition; measures conversational EQ, not task performance.
Higher is better. Elo rating from pairwise comparisons. Typical range: 900-1400.
Judge model (Claude 3.7 Sonnet) may have biases.
| # | Model | Score |
|---|---|---|
| 1 | 1648.0 | |
| 2 | 1594.0 | |
| 3 | 1586.0 | |
| 4 | 1586.0 | |
| 5 | 1586.0 | |
| 6 | 1586.0 | |
| 7 | 1566.0 | |
| 8 | 1549.0 | |
| 9 | 1549.0 | |
| 10 | 1546.0 | |
| 11 | 1533.0 | |
| 12 | 1519.0 | |
| 13 | 1484.0 | |
| 14 | 1461.0 | |
| 15 | 1460.0 | |
| 16 | 1427.0 | |
| 17 | 1427.0 | |
| 18 | 1416.0 | |
| 19 | 1413.0 | |
| 20 | 1384.0 | |
| 21 | 1383.0 | |
| 22 | 1382.0 | |
| 23 | 1330.0 | |
| 24 | 1328.0 | |
| 25 | 1328.0 | |
| 26 | 1319.0 | |
| 27 | 1318.0 | |
| 28 | 1318.0 | |
| 29 | 1314.0 | |
| 30 | 1314.0 | |
| 31 | 1311.0 | |
| 32 | 1311.0 | |
| 33 | 1310.0 | |
| 34 | 1308.0 | |
| 35 | 1308.0 | |
| 36 | 1307.0 | |
| 37 | 1307.0 | |
| 38 | 1303.0 | |
| 39 | 1302.0 | |
| 40 | 1293.0 | |
| 41 | 1293.0 | |
| 42 | 1288.0 | |
| 43 | 1287.0 | |
| 44 | 1281.0 | |
| 45 | 1271.0 | |
| 46 | 1267.0 | |
| 47 | 1267.0 | |
| 48 | 1264.0 | |
| 49 | 1263.0 | |
| 50 | 1260.0 |