KoGPT is Kakao Brain’s Korean Generative Pre-trained Transformer, an open GPT-style language model built specifically for the Korean language. Published on GitHub with about 1,011 stars, the project releases the KoGPT-6B model โ roughly 6.2 billion parameters โ together with the code to use it, and the weights are distributed through Hugging Face. The repository was last updated in 2024 and remains a reference point for Korean NLP research.
When KoGPT appeared, large generative models were overwhelmingly English-first, and Korean needed a capable open alternative. KoGPT fills that gap with a decoder-only transformer in the GPT-3 style, pretrained on large-scale Korean text. The GitHub repository documents the model family, including the KoGPT6B-ryan1.5b checkpoints in full and half precision, and links to the corresponding Hugging Face releases so researchers can download and run the model directly. For teams building Korean chatbots, summarizers, or generation features in the era before today’s multilingual frontier models, KoGPT provided a strong, openly available foundation.
Using KoGPT is deliberately familiar to anyone who has worked with Hugging Face: the models load through the standard AutoTokenizer and AutoModelForCausalLM classes, so Korean text generation plugs into the usual Transformers pipelines with a few lines of code. Kakao Brain released the project openly so Korean developers would not have to start from English-only checkpoints, and the repository documents the released checkpoints and their configurations in detail for researchers who want to reproduce or extend the work.
KoGPT is free. The code is released under Apache-2.0, while the model weights are distributed under a CC BY-NC-ND 4.0 license, which limits them to non-commercial use.
Korean NLP researchers and developers who need an openly available Korean generative model, whether for academic study, benchmarking, or building Korean-language applications.