Commercial training data from specialist providers. Listings show what each source advertises; licensing is confirmed with the provider before any agreement.
This dataset contains images of tables in Korean, captured using smartphones or digital cameras.
This dataset contains images of whiteboards and blackboards in Korean, captured using smartphones or digital cameras.
A low-resource language parallel corpus dataset built by translating Korean written and spoken-style source texts into eight low-resource languages, including…
A multilingual AI model evaluation dataset built by having expert reviewers re-evaluate and verify automatically scored response quality according to defined evaluation…
A multilingual QA benchmark dataset built for evaluating rapidly changing knowledge and factual accuracy across general domains, curated by professional annotators.
A multilingual instruction-following QA text dataset built for tasks that require generating answers based on referenced documents.
A multilingual food and menu parallel translation corpus built by translating menu names extracted from sources such as AI Hub, the Korea Tourism Organization, and…
A multilingual general-domain QA text dataset covering topics such as politics, finance, and society, reviewed and refined by professional annotators.
A multilingual OCR image dataset for menu images in seven languages, including Korean, English, Chinese, and Japanese, comprising original images, translated-text…
A multilingual multi-turn chat text dataset built directly by professional annotators, including speaker information, proper nouns, chat slang, typo tags, and natural…
A multilingual STEM preference evaluation dataset built from STEM-domain questions and three AI-generated answers, with STEM majors providing scoring, feedback, and…
A multilingual STEM QA text dataset built from expert-level questions and answers in science, technology, engineering, and mathematics, reviewed and refined by STEM…
A dataset consisting of AI-generated videos sampled at 1fps with frame-level scene description captions.
A code-switching multi-turn spoken dialogue dataset in which two or more languages are naturally mixed within a single conversation.
A dataset of short voice commands and warning turns used in emergency situations such as disasters, accidents, security incidents, medical emergencies, and defense…
A parallel speech dataset consisting of sentence-level aligned English and Korean utterances, designed for training Speech-to-Speech translation models.
A CoT dataset designed to train contextual understanding and logical step-by-step reasoning based on game press releases, reporting, and interview articles.
A raw text dataset broadly collecting press releases, news updates, field reporting, and interview articles from game media.
A Korean-based Arena benchmark text dataset built for evaluating AI model performance in general-domain comparison tasks.
A Korean-based BFCL 3v benchmark text dataset built for evaluating function-calling capabilities of AI models.
A Korean-based BFCL 4v benchmark text dataset built for evaluating function-calling capabilities of AI models.
A high-difficulty Korean reasoning benchmark dataset built according to global HLE standards, including complex expert-level reasoning problems.
A Korean-based long context benchmark text dataset built to evaluate AI models' ability to process and reason over extended contexts.
A Korean-based Chat Arena preference evaluation multi turn dataset built by collecting human preferences for AI model responses.