# EQ-Bench

Tests how well AI models handle emotionally sensitive conversations with simulated people who have different personalities and needs.

Category: reasoning. Type: leaderboard. Status: active.
Metric: EQ-Bench 4 Elo rating (higher means judges preferred the model more often) (higher is better).
Source: EQ-Bench (Sam Paech) — https://eqbench.com/eqbench4/eqbench4_data.js, results as of 2026-10-24, retrieved 2026-11-17, license: not stated.

| Rank | Model | Elo | 95% CI |
|---|---|---|---|
| 1 | claude-opus-5 | 1364.9 | 1349.5 - 1378.4 |
| 2 | claude-opus-5-5 | 1339.5 | 1329.4 - 1349.6 |
| 3 | muse-wonder | 1324.9 | 1312.5 - 1339.0 |
| 4 | kimi-k3 | 1319.2 | 1308.7 - 1333.0 |
| 5 | claude-fable-5 | 1312.5 | 1302.8 - 1323.6 |
| 6 | gpt-5.5 | 1295.4 | 1285.0 - 1305.6 |
| 7 | claude-opus-4-7 | 1293.2 | 1283.6 - 1303.4 |
| 8 | gpt-6.1-sol | 1272.4 | 1262.9 - 1282.7 |
| 9 | claude-opus-4-8 | 1267.8 | 1258.6 - 1277.0 |
| 10 | gpt-5.4 | 1257.7 | 1250.5 - 1265.3 |
| 11 | muse-spark-1.1 | 1244.4 | 1230.8 - 1256.2 |
| 12 | gemini-4-argon | 1236.3 | 1225.0 - 1248.1 |
| 13 | gpt-5.6-sol | 1236.1 | 1225.4 - 1247.7 |
| 14 | claude-sonnet-5 | 1221.2 | 1211.6 - 1233.5 |
| 15 | gpt-5.6-terra | 1220.1 | 1209.9 - 1229.9 |
| 16 | Inkling | 1212.1 | 1201.2 - 1222.0 |
| 17 | claude-opus-4-6 | 1210.0 | 1198.5 - 1219.6 |
| 18 | glm-5.2 | 1208.1 | 1197.1 - 1216.9 |
| 19 | mimo-v2.5-pro | 1197.7 | 1185.0 - 1212.5 |
| 20 | claude-sonnet-4-6 | 1196.8 | 1187.9 - 1204.9 |
| 21 | kimi-k2.6 | 1191.4 | 1181.4 - 1200.8 |
| 22 | deepseek-v4-pro | 1156.3 | 1144.7 - 1166.7 |
| 23 | gpt-5.6-luna | 1145.9 | 1134.2 - 1156.6 |
| 24 | minimax-m3 | 1140.1 | 1128.8 - 1152.1 |
| 25 | gemini-3.1-pro-preview | 1132.3 | 1118.9 - 1144.4 |

Notes: EQ-Bench 4 (120 personas, judged by Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.6); sorted by Elo from site data file. Date is data generation time.
