Skip to content

Benchmark

Horangi 4 subset: ko_mtbench

People compare model answers and choose which they prefer. Higher scores are better.

Models measured
79
Scores recorded
83
Sources
1
IQ weight
1.76

Score distribution

0.085
Mean
0.087
Median
0.009
Std dev
0.081
10th percentile
0.089
90th percentile
0.0234
Min
0.02340.0901
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.76 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
1Claude Opus 4.6claude-opus-4-6: high-effort· Horangi Korean Leaderboard 4 data docs0.0901
2Claude Opus 4.5claude-opus-4-5-20251101: high-effort· Horangi Korean Leaderboard 4 data docs0.0896
3Gemini 3.5 Flashgemini-3.5-flash· Horangi Korean Leaderboard 4 data docs0.0896
4GPT 5.1gpt-5.1-2025-11-13· Horangi Korean Leaderboard 4 data docs0.0896
5Gemini 3 Flashgemini-3-flash-preview· Horangi Korean Leaderboard 4 data docs0.0896
6Claude Sonnet 4.6claude-sonnet-4-6: high-effort· Horangi Korean Leaderboard 4 data docs0.0895
7Claude Sonnet 4.5claude-sonnet-4-5-20250929: high-effort· Horangi Korean Leaderboard 4 data docs0.0894
8Claude Sonnet 5claude-sonnet-5: high-effort· Horangi Korean Leaderboard 4 data docs0.0893
9Gemini 3.1 Pro Previewgemini-3.1-pro-preview: high-effort· Horangi Korean Leaderboard 4 data docs0.0893
10GPT OSS 120Bgpt-oss-120b: high-effort· Horangi Korean Leaderboard 4 data docs0.0893
11Gemini 3 Pro Previewgemini-3-pro-preview: high-effort· Horangi Korean Leaderboard 4 data docs0.0891
12Qwen3.7 Maxqwen3.7-max· Horangi Korean Leaderboard 4 data docs0.0891
13Gemini 3.1 Flash Litegemini-3.1-flash-lite-preview· Horangi Korean Leaderboard 4 data docs0.089
14MiMo V2 Promimo-v2-pro· Horangi Korean Leaderboard 4 data docs0.089
15Claude Opus 4.1claude-opus-4-1-20250805· Horangi Korean Leaderboard 4 data docs0.0889
16GPT 5.2gpt-5.2-2025-12-11: xhigh-effort· Horangi Korean Leaderboard 4 data docs0.0889
17Qwen3.7 Plusqwen3.7-plus· Horangi Korean Leaderboard 4 data docs0.0887
18Gemma 4 31B ITgemma-4-31B-it: enable-thinking· Horangi Korean Leaderboard 4 data docs0.0886
19GLM 4.7glm-4.7· Horangi Korean Leaderboard 4 data docs0.0886
20Kimi K2.5kimi-k2.5· Horangi Korean Leaderboard 4 data docs0.0886
21Claude Haiku 4.5claude-haiku-4-5-20251001: high-effort· Horangi Korean Leaderboard 4 data docs0.0885
22Gemma 4 E4B-itgemma-4-E4B-it: enable-thinking· Horangi Korean Leaderboard 4 data docs0.0883
23GPT 4.1gpt-4.1· Horangi Korean Leaderboard 4 data docs0.0883
24GPT-5.6 Lunagpt-5.6-luna: max-effort· Horangi Korean Leaderboard 4 data docs0.0883
25Claude Opus 4.8claude-opus-4-8: high-effort· Horangi Korean Leaderboard 4 data docs0.0883
26GLM 5glm-5· Horangi Korean Leaderboard 4 data docs0.088
27Claude Opus 4claude-opus-4-20250514: high-effort· Horangi Korean Leaderboard 4 data docs0.0879
28Claude Fable 5claude-fable-5: high-effort· Horangi Korean Leaderboard 4 data docs0.0876
29Qwen 3 235B A22B Thinking 2507qwen3-235b-a22b-thinking-2507· Horangi Korean Leaderboard 4 data docs0.0875
30GPT 5.5gpt-5.5: xhigh-effort· Horangi Korean Leaderboard 4 data docs0.0874
31Qwen3.6 Plus Previewqwen3.6-plus-preview· Horangi Korean Leaderboard 4 data docs0.0873
32Gemma 4 31B ITGemma-4-31B-IT-NVFP4: enable-thinking· Horangi Korean Leaderboard 4 data docs0.0873
33GPT-5.4 minigpt-5.4-mini-2026-03-17: xhigh-effort· Horangi Korean Leaderboard 4 data docs0.0873
34Grok 4.20grok-4.20-beta· Horangi Korean Leaderboard 4 data docs0.0873
35Qwen3.5 122B-A10Bqwen3.5-122b-a10b· Horangi Korean Leaderboard 4 data docs0.0873
36DeepSeek Reasonerdeepseek-v3.2: high-effort· Horangi Korean Leaderboard 4 data docs0.0873
37GPT 5.4 Nanogpt-5.4-nano-2026-03-17: xhigh-effort· Horangi Korean Leaderboard 4 data docs0.0871
38GPT-5.6 Terragpt-5.6-terra: max-effort· Horangi Korean Leaderboard 4 data docs0.087
39GLM 5V Turboglm-5v-turbo· Horangi Korean Leaderboard 4 data docs0.087
40Gemma 4 E2B-itgemma-4-E2B-it: enable-thinking· Horangi Korean Leaderboard 4 data docs0.0869
41GLM 4.5 Airglm-4.5-air· Horangi Korean Leaderboard 4 data docs0.0869
42GLM 5.2glm-5.2: high-effort· Horangi Korean Leaderboard 4 data docs0.0869
43Qwen3.5 397B A17Bqwen3.5-397b-a17b· Horangi Korean Leaderboard 4 data docs0.0868
44GLM 5 Turboglm-5-turbo· Horangi Korean Leaderboard 4 data docs0.0866
45GPT 5.4gpt-5.4-2026-03-05: xhigh-effort· Horangi Korean Leaderboard 4 data docs0.0866
46Grok 4.20grok-4.20· Horangi Korean Leaderboard 4 data docs0.0866
47Gemma 4 26B A4Bgemma-4-26B-A4B-it: enable-thinking· Horangi Korean Leaderboard 4 data docs0.0865
48GPT 5gpt-5-2025-08-07· Horangi Korean Leaderboard 4 data docs0.0865
49GPT-5 minigpt-5-mini-2025-08-07· Horangi Korean Leaderboard 4 data docs0.0865
50GPT 4ogpt-4o-2024-11-20· Horangi Korean Leaderboard 4 data docs0.0864

Showing 1–50 of 83 results.

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