Skip to content

Benchmark

Horangi 4 subset: kmmlu_pro

Higher scores are better.

Models measured
79
Scores recorded
83
Sources
1
IQ weight
1.93

Score distribution

0.7
Mean
0.74
Median
0.173
Std dev
0.5
10th percentile
0.9
90th percentile
0.13
Min
0.130.95
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.93 for chat models, measured over 79 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51MiMo V2 Flashmimo-v2-flash· Horangi Korean Leaderboard 4 data docs0.671
52GPT 4.1gpt-4.1· Horangi Korean Leaderboard 4 data docs0.67
53Grok 4.1 Fastgrok-4-1-fast-reasoning· Horangi Korean Leaderboard 4 data docs0.67
54Qwen3 14BQwen3-14B· Horangi Korean Leaderboard 4 data docs0.67
55MiniMax M2.5minimax-m2.5· Horangi Korean Leaderboard 4 data docs0.663
56Nemotron 3 Super 120B A12BNVIDIA-Nemotron-3-Super-120B-A12B-FP8· Horangi Korean Leaderboard 4 data docs0.66
57Qwen3.5 9Bqwen3.5-9b· Horangi Korean Leaderboard 4 data docs0.642
58Claude Haiku 4.5claude-haiku-4-5-20251001: high-effort· Horangi Korean Leaderboard 4 data docs0.64
59GPT OSS 120Bgpt-oss-120b: high-effort· Horangi Korean Leaderboard 4 data docs0.64
60GPT-5 nanogpt-5-nano-2025-08-07· Horangi Korean Leaderboard 4 data docs0.63
61Qwen3 32BQwen3-32B· Horangi Korean Leaderboard 4 data docs0.62
62MiniMax M2.7minimax-m2.7· Horangi Korean Leaderboard 4 data docs0.616
63Mistral Medium 3.5mistral-medium-3-5· Horangi Korean Leaderboard 4 data docs0.602
64GPT-4.1 minigpt-4.1-mini· Horangi Korean Leaderboard 4 data docs0.6
65GLM 4.5 Airglm-4.5-air· Horangi Korean Leaderboard 4 data docs0.583
66Qwen3 4BQwen3-4B· Horangi Korean Leaderboard 4 data docs0.55
67GLM 4.7 Flashglm-4.7-flash· Horangi Korean Leaderboard 4 data docs0.545
68Gemma 4 E4B-itgemma-4-E4B-it: enable-thinking· Horangi Korean Leaderboard 4 data docs0.54
69Grok 4 Fastgrok-4-fast-non-reasoning· Horangi Korean Leaderboard 4 data docs0.54
70Grok 4.1 Fastgrok-4-1-fast-non-reasoning· Horangi Korean Leaderboard 4 data docs0.53
71Gemma 3 27B ITgemma-3-27b-it· Horangi Korean Leaderboard 4 data docs0.52
72Qwen3 8BQwen3-8B· Horangi Korean Leaderboard 4 data docs0.52
73Nemotron 3 Nano 30B A3BNVIDIA-Nemotron-3-Nano-30B-A3B-BF16· Horangi Korean Leaderboard 4 data docs0.51
74Gemma 4 E2B-itgemma-4-E2B-it: enable-thinking· Horangi Korean Leaderboard 4 data docs0.5
75Mistral Smallmistral-small-2603· Horangi Korean Leaderboard 4 data docs0.5
76Qwen3 4B Instruct 2507Qwen3-4B-Instruct-2507· Horangi Korean Leaderboard 4 data docs0.48
77GPT-4.1 nanogpt-4.1-nano· Horangi Korean Leaderboard 4 data docs0.46
78Gemma 3 12B ITgemma-3-12b-it· Horangi Korean Leaderboard 4 data docs0.45
79Qwen3 1.7BQwen3-1.7B· Horangi Korean Leaderboard 4 data docs0.41
80Gemma 3 4B ITgemma-3-4b-it· Horangi Korean Leaderboard 4 data docs0.32
81Qwen3 0.6BQwen3-0.6B· Horangi Korean Leaderboard 4 data docs0.26
82Gemma 3 1b itgemma-3-1b-it· Horangi Korean Leaderboard 4 data docs0.16
83Gemma 3 270M Itgemma-3-270m-it· Horangi Korean Leaderboard 4 data docs0.13

Showing 51–83 of 83 results.

Data credits: Horangi 4, Weights & Biases Korea; public publisher-selected numeric evaluation results