# Korean > Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/scale%3Akorean. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/scale%3Akorean. The Scale AI Multilingual Prompts Dataset, composed of 1,000 prompts per language, is tailored to enhance models’ interaction capabilities across multiple languages. This dataset specifically aims to refine chatbots' proficiency in engaging with Korean users, reflecting complexity of Korean communication. This dataset introduces single-turn prompts across a diversified range of scenarios, aiming to evaluate and improve models’ responses in both general and culturally nuanced conversations. Canonical URL: https://benchize.com/benchmarks/scale%3Akorean Last updated: Wed, 21 Oct 2026 18:52:20 GMT Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations. The Scale AI Multilingual Prompts Dataset, composed of 1,000 prompts per language, is tailored to enhance models’ interaction capabilities across multiple languages. This dataset specifically aims to refine chatbots' proficiency in engaging with Korean users, reflecting complexity of Korean communication. This dataset introduces single-turn prompts across a diversified range of scenarios, aiming to evaluate and improve models’ responses in both general and culturally nuanced conversations. Language & multilingual · Models tested: 8 · Score publishers: [scale](https://labs.scale.com/leaderboard/korean) ## Model results | # | Model | Score | ± | | | --- | --- | --- | --- | --- | | 1 | [GPT-4 Turbo Preview](https://benchize.com/models/gpt-4-turbo-preview.md) | 60.76source score | CI 55.12–66.4 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 60.76 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration GPT-4 Turbo Preview sourceModelVersion gpt-4-0125-preview thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 55.12–66.4 contaminationWarning Not reported rank 5 reportedConfidenceHalfWidth 5.64 ### Source [Original source](https://labs.scale.com/leaderboard/korean) Source scale Original model label GPT-4 Turbo Preview Observed 2026-10-21T16:35:44.172Z Evidence SHA-256 `e1846e496288a3d1d0a49f645276e266d451f5bb5704de0e8e1a771e011f803b` Source locator `/rows/4` | | | | | | 2 | [Gemini 2.0 Flash](https://benchize.com/models/gemini-2-0-flash.md) | 55.6source score | CI 49.87–61.33 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 55.6 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration Gemini 2.0 Flash sourceModelVersion Not reported thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 49.87–61.33 contaminationWarning Not reported rank 9 reportedConfidenceHalfWidth 5.73 ### Source [Original source](https://labs.scale.com/leaderboard/korean) Source scale Original model label Gemini 2.0 Flash Observed 2026-10-21T16:35:44.172Z Evidence SHA-256 `e1846e496288a3d1d0a49f645276e266d451f5bb5704de0e8e1a771e011f803b` Source locator `/rows/8` | | | | | | 3 | [Claude 3 Sonnet](https://benchize.com/models/claude-3-sonnet.md) | 54.17source score | CI 48.39–59.95 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 54.17 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration Claude 3 Sonnet sourceModelVersion claude-3-sonnet-20240229 thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 48.39–59.95 contaminationWarning Not reported rank 10 reportedConfidenceHalfWidth 5.78 ### Source [Original source](https://labs.scale.com/leaderboard/korean) Source scale Original model label Claude 3 Sonnet Observed 2026-10-21T16:35:44.172Z Evidence SHA-256 `e1846e496288a3d1d0a49f645276e266d451f5bb5704de0e8e1a771e011f803b` Source locator `/rows/9` | | | | | | 4 | [Claude 3 Opus](https://benchize.com/models/claude-3-opus.md) | 52.78source score | CI 47.01–58.55 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 52.78 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration Claude 3 Opus sourceModelVersion claude-3-opus-20240229 thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 47.01–58.55 contaminationWarning Not reported rank 11 reportedConfidenceHalfWidth 5.77 ### Source [Original source](https://labs.scale.com/leaderboard/korean) Source scale Original model label Claude 3 Opus Observed 2026-10-21T16:35:44.172Z Evidence SHA-256 `e1846e496288a3d1d0a49f645276e266d451f5bb5704de0e8e1a771e011f803b` Source locator `/rows/10` | | | | | | 5 | [GPT-4o Mini](https://benchize.com/models/gpt-4o-mini.md) | 51.74source score | CI 45.97–57.51 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 51.74 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration GPT-4o mini sourceModelVersion gpt-4o-mini-2024-07-18 thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 45.97–57.51 contaminationWarning Not reported rank 12 reportedConfidenceHalfWidth 5.77 ### Source [Original source](https://labs.scale.com/leaderboard/korean) Source scale Original model label GPT-4o mini Observed 2026-10-21T16:35:44.172Z Evidence SHA-256 `e1846e496288a3d1d0a49f645276e266d451f5bb5704de0e8e1a771e011f803b` Source locator `/rows/11` | | | | | | 6 | [GPT-4](https://benchize.com/models/gpt-4.md) | 51.39source score | CI 45.62–57.16 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 51.39 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration GPT-4 sourceModelVersion gpt-4-0613 thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 45.62–57.16 contaminationWarning Not reported rank 13 reportedConfidenceHalfWidth 5.77 ### Source [Original source](https://labs.scale.com/leaderboard/korean) Source scale Original model label GPT-4 Observed 2026-10-21T16:35:44.172Z Evidence SHA-256 `e1846e496288a3d1d0a49f645276e266d451f5bb5704de0e8e1a771e011f803b` Source locator `/rows/12` | | | | | | 7 | [Llama 3.1 70B Instruct](https://benchize.com/models/llama-3-1-70b-instruct.md) | 37.23source score | CI 31.63–42.83 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 37.23 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration Llama 3.1 70B Instruct sourceModelVersion llama-3.1-70b-instruct thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 31.63–42.83 contaminationWarning Not reported rank 17 reportedConfidenceHalfWidth 5.6 ### Source [Original source](https://labs.scale.com/leaderboard/korean) Source scale Original model label Llama 3.1 70B Instruct Observed 2026-10-21T16:35:44.172Z Evidence SHA-256 `e1846e496288a3d1d0a49f645276e266d451f5bb5704de0e8e1a771e011f803b` Source locator `/rows/16` | | | | | | 8 | [Llama 3.1 8B Instruct](https://benchize.com/models/llama-3-1-8b-instruct.md) | 17.42source score | CI 13.03–21.81 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 17.42 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration Llama 3.1 8B Instruct sourceModelVersion llama-3.1-8b-instruct thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 13.03–21.81 contaminationWarning Not reported rank 19 reportedConfidenceHalfWidth 4.39 ### Source [Original source](https://labs.scale.com/leaderboard/korean) Source scale Original model label Llama 3.1 8B Instruct Observed 2026-10-21T16:35:44.172Z Evidence SHA-256 `e1846e496288a3d1d0a49f645276e266d451f5bb5704de0e8e1a771e011f803b` Source locator `/rows/18` | | | | | --- Cite as: Korean — Benchize, https://benchize.com/benchmarks/scale%3Akorean (retrieved 2026-10-21)