BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

EQ-Bench 4

Emotional intelligence in multi-turn role-play, judged pairwise into an Elo-style rating.

As of 24 Oct 2026, Claude Opus 5 leads EQ-Bench 4 on BenchLeader with 1385, ahead of Claude Fable 5 at 1340, across 27 model configurations with a published result.

Published by
EQ-Bench
Category
Human preference
Index weight
0.5
Models
27
Data as of
24 Oct 2026
Newest result
26 Jul 2026

The publisher has posted no new results on this board since 26 Jul 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.

MIT — EQ-Bench.

What the test looks like

EQ-Bench 4 puts models into multi-turn role-play scenarios that call for emotional intelligence, such as de-escalating a conflict, and has a judge model compare transcripts pairwise.

How it is scored

An Elo-style rating from the pairwise judgments, published by the benchmark's author.

What to keep in mind

Judged by another model, so it inherits that judge's taste, and role-play skill is not the same as being a good assistant. The rating scale is relative to the models entered.

Top 15
  1. 1Claude Opus 51385
  2. 2Claude Fable 51340
  3. 3Kimi K31339
  4. 4GPT-5.51315
  5. 5Claude Opus 4.71311
  6. 6Claude Opus 4.81281
  7. 7GPT-5.41272
  8. 8Muse Spark 1.11260
  9. 9GPT-5.6 Sol1250
  10. 10Claude Sonnet 51236
  11. 11GPT-5.6 Terra1234
  12. 12Claude Opus 4.61223
  13. 13GLM 5.21222
  14. 14MiMo-V2.5-Pro1208
  15. 15Claude Sonnet 4.61207
27 of 27
#
1Claude Opus 5138565.52026-07-26
2Claude Fable 5134067.12026-07-26
3Kimi K3133959.12026-07-26
4GPT-5.5131560.32026-07-26
5Claude Opus 4.7131163.32026-07-26
6Claude Opus 4.8128160.12026-07-26
7GPT-5.4127258.22026-07-26
8Muse Spark 1.1126062.52026-07-26
9GPT-5.6 Sol125055.92026-07-26
10Claude Sonnet 5123656.12026-07-26
11GPT-5.6 Terra123452.72026-07-26
12Claude Opus 4.6122362.62026-07-26
13GLM 5.2122252.02026-07-26
14MiMo-V2.5-Pro120854.92026-07-26
15Claude Sonnet 4.6120757.12026-07-26
16Kimi K2.6120255.42026-07-26
17DeepSeek V4 Pro116653.52026-07-26
18GPT-5.6 Luna115648.02026-07-26
19MiniMax M3115056.42026-07-26
20Gemini 3.1 Pro114260.62026-07-26
21Gemma 4 31B112048.42026-07-26
22Qwen3.7max111060.62026-07-26
23Gemini 3.5 Flash108749.12026-07-26
24Grok 4.3107554.72026-07-26
25Claude Haiku 4.5106445.42026-07-26
26Qwen3.6 27B102648.62026-07-26
27Mistral Medium 3.599349.42026-07-26

More human preference benchmarks · All LLM benchmarks

Cite as: BenchLeader, “EQ-Bench 4 leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/eqbench4.

EQ-Bench 4: questions

What does EQ-Bench 4 measure?
EQ-Bench 4 puts models into multi-turn role-play scenarios that call for emotional intelligence, such as de-escalating a conflict, and has a judge model compare transcripts pairwise. Scores are reported in Elo-style ratings from pairwise comparisons; higher is better.
Which AI model leads EQ-Bench 4?
Claude Opus 5 leads EQ-Bench 4 with 1385 as of 24 Oct 2026, ahead of Claude Fable 5 at 1340.
How many models have EQ-Bench 4 results?
27 model configurations have a EQ-Bench 4 result on BenchLeader, all taken from EQ-Bench.
Who runs EQ-Bench 4 and how often is it updated?
EQ-Bench 4 is published by EQ-Bench. BenchLeader re-reads the published results once a day and records the date each result was published.
Does EQ-Bench 4 count toward the BenchLeader Index?
Yes. EQ-Bench 4 contributes to the human preference category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.