Skip to content

Benchmark

Judgemark v4

Scores are percentages.

Scores from EQ-Bench boards.
Models measured
42
Scores recorded
42
Sources
1
IQ weight
1.69

Score distribution

56.7%
Mean
57.9%
Median
20.6%
Std dev
28%
10th percentile
82.1%
90th percentile
15.5%
Min
15.5%90.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.69 for chat models, measured over 42 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 4.6· EQ-Bench boards90.7%
2GPT 5.5· EQ-Bench boards87.8%
3Claude Opus 4.7· EQ-Bench boards84%
4Gemini 3.7 Flash· EQ-Bench boards83.4%
5Claude Sonnet 4.6· EQ-Bench boards82.1%
6Muse Spark 1.2· EQ-Bench boards79.4%
7Claude Opus 5· EQ-Bench boards78.8%
8Gemini 3.1 Pro Preview· EQ-Bench boards78.7%
9Claude Opus 4.8· EQ-Bench boards78%
10Grok 4.5· EQ-Bench boards77.1%
11GLM 5.2· EQ-Bench boards73.2%
12GLM 5.3· EQ-Bench boards72.6%
13Gemma 4 31B IT· EQ-Bench boards72.3%
14GPT 5.4· EQ-Bench boards72.1%
15Claude Sonnet 5· EQ-Bench boards70.9%
16Gemini 3.5 Flash· EQ-Bench boards68.8%
17Qwen 3.8 27B· EQ-Bench boards67.4%
18GLM 5.1· EQ-Bench boards67.2%
19Qwen3.5 27B· EQ-Bench boards60.5%
20Gemini 3.1 Flash Lite· EQ-Bench boards58.8%
21MiMo V2.5 Pro· EQ-Bench boards57.9%
22Kimi K2.6· EQ-Bench boards57.3%
23Qwen3.5 397B A17B· EQ-Bench boards54.2%
24Qwen3.6 27B· EQ-Bench boards53.4%
25Gemma 4 26B A4B· EQ-Bench boards53%
26Nemotron 3 Ultra· EQ-Bench boards52.9%
27Grok 4.3· EQ-Bench boards49.6%
28DeepSeek V4 Pro· EQ-Bench boards47.1%
29Gemini 3 Flash· EQ-Bench boards46.1%
30Qwen3.6 Max Preview· EQ-Bench boards45.1%
31Muse Glimmer 30B· EQ-Bench boards40.7%
32Qwen3.5 35B A3B· EQ-Bench boards40.5%
33DeepSeek V4 Flash· EQ-Bench boards36.8%
34Qwen3.6 35B A3B· EQ-Bench boards32.7%
35Qwen3.5 9B· EQ-Bench boards32.4%
36Qwen3.6 Flash· EQ-Bench boards31.2%
37DeepSeek V4 Flash 0731· EQ-Bench boards30.7%
38Step 3.7 Flash· EQ-Bench boards28%
39GPT 5.4 Nano· EQ-Bench boards27.5%
40GPT-5.4 mini· EQ-Bench boards27.3%
41Mistral Small 4 119B· EQ-Bench boards16.5%
42GPT OSS 120B· EQ-Bench boards15.5%

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