Skip to content

Benchmark

Longform Creative Writing

Scores are percentages.

Scores from EQ-Bench boards.
Models measured
124
Scores recorded
124
Sources
1
IQ weight
1.57

Score distribution

59.2%
Mean
60.2%
Median
17.4%
Std dev
32.9%
10th percentile
81.5%
90th percentile
17%
Min
17%86.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.57 for chat models, measured over 124 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51GPT-5 Chat· EQ-Bench boards66.8%
52Qwen3 Max· EQ-Bench boards66.2%
53DeepSeek V4 Flash· EQ-Bench boards66%
54GLM 4.7· EQ-Bench boards66%
55MiniMax M2.5· EQ-Bench boards65.3%
56Claude Haiku 4.5· EQ-Bench boards65%
57OpenAI o3· EQ-Bench boards62.7%
58GPT-5.3 Chat· EQ-Bench boards62.6%
59DeepSeek V3.1· EQ-Bench boards62.3%
60DeepSeek V4 Flash 0731· EQ-Bench boards61.2%
61ChatGPT-4o Latest· EQ-Bench boards60.2%
62DeepSeek V3.2 Exp· EQ-Bench boards60.2%
63DeepSeek V3· EQ-Bench boards60.1%
64DeepSeek R1· EQ-Bench boards59.5%
65Qwen3.5 27B· EQ-Bench boards59%
66Gemini 2.5 Pro Preview 03-25· EQ-Bench boards58.8%
67Claude 3.5 Sonnet· EQ-Bench boards58.4%
68Claude Sonnet 4· EQ-Bench boards58.4%
69GPT 5.4 Nano· EQ-Bench boards57.8%
70MiniMax M2.1· EQ-Bench boards57.8%
71Claude Sonnet 3.7· EQ-Bench boards57.7%
72Qwen3 235B A22B Instruct 2507· EQ-Bench boards57.4%
73GLM 4.6· EQ-Bench boards57.3%
74Gemma 4 31B IT· EQ-Bench boards56.5%
75GPT-5 mini· EQ-Bench boards56.2%
76Ling 1t· EQ-Bench boards56.2%
77Gemini 3 Flash· EQ-Bench boards55.7%
78GLM 4.5· EQ-Bench boards55.5%
79Qwen3.5 122B-A10B· EQ-Bench boards54.3%
80Gemini 2.5 Pro· EQ-Bench boards53.6%
81Qwen 3.8 27B· EQ-Bench boards53.3%
82MiMo V2 Flash· EQ-Bench boards53.1%
83Hermes 4 405B· EQ-Bench boards51.2%
84Gemma 4 26B A4B· EQ-Bench boards50.7%
85Gemini 3.1 Flash Lite· EQ-Bench boards49%
86DeepSeek V3.2 Speciale· EQ-Bench boards48.8%
87Gemma 4 12B It· EQ-Bench boards48.7%
88Gemini 2.5 Flash Preview 05-20 (no think)· EQ-Bench boards48.6%
89Mistral Large 3 675b Instruct· EQ-Bench boards48.3%
90GLM 4.7 Flash· EQ-Bench boards47.8%
91Doubao-Seed 1.6· EQ-Bench boards47.3%
92Olmo 3.1 32B Think· EQ-Bench boards47%
93MiniMax M2· EQ-Bench boards46.4%
94Mistral Medium 3.1· EQ-Bench boards46.2%
95Grok 4· EQ-Bench boards45.7%
96Qwen3.5 35B A3B· EQ-Bench boards44.5%
97Mistral Small 4 119B· EQ-Bench boards43.2%
98Gemma 3 27B IT· EQ-Bench boards41.9%
99Gemma 3 12B IT· EQ-Bench boards41.8%
100Mistral Small 3.2 24B Instruct· EQ-Bench boards41.6%

Showing 51–100 of 124 results.

Data credits: EQ-Bench (Samuel J. Paech) — cite paech2023eqbench