Benchmark Almanac

EQ-Bench 4

Claude Opus 5 1,365 leads since Jul 2026 Not in model cards

Models hold 16-message roleplay chats with 120 simulated user personas whose preferences compete and are hidden. A three-judge panel compares transcripts pairwise, and ratings are fitted as Elo scores.

Gemini 3.1 Pro Preview plays the persona. Claude Opus 4.6, GPT-5.5 and Gemini 3.1 Pro Preview judge. It went live on eqbench.com on 2026-07-23 and moved EQ-Bench 3 to the legacy boards.

Made by EQ-Bench (Sam Paech) Released Jul 2026 Multi-turn conversation All tasks public Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
EQ-bench/eqbench4 9 stars Leaderboard eqbench.com
32
Models scored on its own leaderboard

Claude Opus 5 leads with 1,365

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best.

32 models. Scroll sideways for the rest. Bars start at 900.

  1. 1,365 Claude Opus 5
  2. 1,340 Claude Opus 5.5
  3. 1,325 Muse Wonder
  4. 1,319 Kimi K3
  5. 1,313 Claude Fable 5
  6. 1,295 GPT-5.5
  7. 1,293 Claude Opus 4.7
  8. 1,272 GPT-6.1 Sol
  9. 1,268 Claude Opus 4.8
  10. 1,258 GPT-5.4
  11. 1,244 Muse Spark 1.1
  12. 1,236 Gemini 4 Argon
  13. 1,236 GPT-5.6 Sol
  14. 1,221 Claude Sonnet 5
  15. 1,220 GPT-5.6 Terra
  16. 1,212 Inkling
  17. 1,210 Claude Opus 4.6
  18. 1,208 GLM-5.2
  19. 1,198 MiMo-V2.5-Pro
  20. 1,197 Claude Sonnet 4.6
  21. 1,191 Kimi K2.6
  22. 1,156 DeepSeek-V4-Pro
  23. 1,146 GPT-5.6 Luna
  24. 1,140 MiniMax-M3
  25. 1,132 Gemini 3.1 Pro
  26. 1,110 Gemma 4 31B IT
  27. 1,100 Qwen3.7-Max
  28. 1,077 Gemini 3.5 Flash
  29. 1,065 Grok 4.3
  30. 1,053 Claude Haiku 4.5
  31. 1,015 Qwen3.6 27B
  32. 983 Mistral Medium 3.5

Bars start at 900.

  1. 1 Claude Opus 5 1,365
  2. 2 Claude Opus 5.5 1,340
  3. 3 Muse Wonder 1,325
  4. 4 Kimi K3 1,319
  5. 5 Claude Fable 5 1,313
  6. 6 GPT-5.5 1,295
  7. 7 Claude Opus 4.7 1,293
  8. 8 GPT-6.1 Sol 1,272
  9. 9 Claude Opus 4.8 1,268
  10. 10 GPT-5.4 1,258
  11. 11 Muse Spark 1.1 1,244
  12. 12 Gemini 4 Argon 1,236
  13. 13 GPT-5.6 Sol 1,236
  14. 14 Claude Sonnet 5 1,221
  15. 15 GPT-5.6 Terra 1,220
  16. 16 Inkling 1,212
  17. 17 Claude Opus 4.6 1,210
  18. 18 GLM-5.2 1,208
  19. 19 MiMo-V2.5-Pro 1,198
  20. 20 Claude Sonnet 4.6 1,197
  21. 21 Kimi K2.6 1,191
  22. 22 DeepSeek-V4-Pro 1,156
  23. 23 GPT-5.6 Luna 1,146
  24. 24 MiniMax-M3 1,140
  25. 25 Gemini 3.1 Pro 1,132
  26. 26 Gemma 4 31B IT 1,110
  27. 27 Qwen3.7-Max 1,100
  28. 28 Gemini 3.5 Flash 1,077
  29. 29 Grok 4.3 1,065
  30. 30 Claude Haiku 4.5 1,053
  31. 31 Qwen3.6 27B 1,015
  32. 32 Mistral Medium 3.5 983
#ModelScoreSettingDateSource
1 Claude Opus 5 1,365 Jul 2026 Official leaderboard
2 Claude Opus 5.5 1,340 Sep 2026 Official leaderboard
3 Muse Wonder 1,325 Oct 2026 Official leaderboard
4 Kimi K3 1,319 Jul 2026 Official leaderboard
5 Claude Fable 5 1,313 Jun 2026 Official leaderboard
6 GPT-5.5 1,295 Apr 2026 Official leaderboard
7 Claude Opus 4.7 1,293 Apr 2026 Official leaderboard
8 GPT-6.1 Sol 1,272 Sep 2026 Official leaderboard
9 Claude Opus 4.8 1,268 May 2026 Official leaderboard
10 GPT-5.4 1,258 Mar 2026 Official leaderboard
11 Muse Spark 1.1 1,244 Jul 2026 Official leaderboard
12 Gemini 4 Argon 1,236 Sep 2026 Official leaderboard
13 GPT-5.6 Sol 1,236 Jul 2026 Official leaderboard
14 Claude Sonnet 5 1,221 Jun 2026 Official leaderboard
15 GPT-5.6 Terra 1,220 Jul 2026 Official leaderboard
16 Inkling 1,212 Jul 2026 Official leaderboard
17 Claude Opus 4.6 1,210 Feb 2026 Official leaderboard
18 GLM-5.2 1,208 Jun 2026 Official leaderboard
19 MiMo-V2.5-Pro 1,198 Apr 2026 Official leaderboard
20 Claude Sonnet 4.6 1,197 Feb 2026 Official leaderboard
21 Kimi K2.6 1,191 Apr 2026 Official leaderboard
22 DeepSeek-V4-Pro 1,156 Apr 2026 Official leaderboard
23 GPT-5.6 Luna 1,146 Jul 2026 Official leaderboard
24 MiniMax-M3 1,140 Jun 2026 Official leaderboard
25 Gemini 3.1 Pro 1,132 Feb 2026 Official leaderboard
26 Gemma 4 31B IT 1,110 Apr 2026 Official leaderboard
27 Qwen3.7-Max 1,100 May 2026 Official leaderboard
28 Gemini 3.5 Flash 1,077 May 2026 Official leaderboard
29 Grok 4.3 1,065 Apr 2026 Official leaderboard
30 Claude Haiku 4.5 1,053 Oct 2025 Official leaderboard
31 Qwen3.6 27B 1,015 Apr 2026 Official leaderboard
32 Mistral Medium 3.5 983 Apr 2026 Official leaderboard

3 models have held number one since its release in Jul 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic Moonshot AI
Own leaderboard 3 leaders
Claude Opus 5 1,365

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard
900 1,000 1,100 1,200 1,300 1,400 2026 900 1,000 1,100 1,200 1,300 1,400 2026

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

Versions

Replaces EQ-Bench 3

More in Instructions and chat

All 76 in Instructions and chat
  • IFEval In use Models answer about 500 prompts containing verifiable instructions such as word limits or required keywords. Code checks each instruction for strict and loose accuracy. Gemma 4 31B IT 98.9%
  • IFBench In use 58 new out-of-domain verifiable output constraints. Scored by programmatic constraint checks (strict or loose). Nemotron-Cascade 2 30B-A3B 82.9%
  • MultiChallenge In use Multi-turn conversations testing instruction retention, inference memory, versioned editing and self-coherence. Graded by LLM judge with rubrics. GPT-6.1 Sol 82%
  • Multi-IF In use Models follow verifiable instructions over three-turn conversations in eight languages, extending IFEval. Scored by instruction accuracy at each turn. LFM3-1.7B 89.1%
  • AdvancedIF In use Models respond to over 1,600 prompts with complex, multi-turn, and system-level instructions. A judge model checks each response against expert-written rubrics. MAI-Thinking-1 85%
  • IFStruct In use Models generate JSON or YAML that must follow a requested schema and phrasing constraints, without constrained decoding. A response passes only if every structural constraint holds. LFM3-1.7B 93.7%