Kakao Brain (KoGPT)
Open-source Korean GPT model from Kakao — the go-to baseline for Korean NLP research since 2021.
Overview
KoGPT is a 6.17-billion-parameter Korean generative pre-trained transformer released by Kakao Brain in November 2021. Built on a GPT-style decoder architecture with Rotary Position Embedding, it was trained on a large Korean-language corpus (the "ryan" dataset) and became the de-facto academic baseline for Korean NLP tasks — achieving 87.8% accuracy on NSMC sentiment classification and 78.0 F1 on YNAT topic classification. The weights are freely downloadable via Hugging Face and GitHub for non-commercial research under CC-BY-NC-ND 4.0.
Kakao Brain, the AI R&D subsidiary that developed KoGPT, was absorbed into Kakao Corp in mid-2024 as part of a broader AI consolidation under a new Chief AI Officer. A planned KoGPT 2.0 (targeting up to 65B parameters) was indefinitely delayed, with Kakao pivoting instead to a strategic partnership with OpenAI that embedded ChatGPT directly into KakaoTalk. As of 2026, KoGPT remains a widely-cited open-source research artefact, but active product development has shifted to Kakao's Kanana AI service.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- CC BY-NC-ND 4.0 weights, Apache 2.0 code
- Korean 6B generative model
- Successor Kanana models now published by Kakao
- Contact [email protected]
- Case-by-case licensing
Use cases
What you can produce with Kakao Brain (KoGPT)
- Downloadable 6.17B-parameter Korean GPT model weights (float32 and float16 variants) via Hugging Face
- Apache 2.0 licensed inference and tokenisation code on GitHub
- 87.8% NSMC sentiment accuracy and 78.0 F1 YNAT topic classification out of the box
- 64,512-token Hangul-optimised vocabulary and tokeniser
- Community Hugging Face Spaces for browser-based demos (11 active)
- Peer-citable academic reference: Kim et al. 2021 KoGPT: KakaoBrain Korean Generative Pre-trained Transformer
- Drop-in HuggingFace Transformers compatibility for existing Korean NLP pipelines
ASEAN Perspective
Kakao Brain (KoGPT) in Southeast Asia
KoGPT is a Korean-first model with limited direct utility for ASEAN languages — its vocabulary and training data are almost exclusively Hangul, and the planned multilingual KoGPT 2.0 (which would have added Bahasa and Vietnamese support) was never released. That said, the Korean diaspora and Korean-content consumption across the Philippines, Indonesia, Malaysia, and Vietnam (driven by K-pop and K-drama) creates genuine demand for Korean NLP tools in the region. For ASEAN-based researchers studying Korean or building Korean-language consumer apps, KoGPT remains a freely accessible, academically credible baseline — though teams needing full ASEAN language coverage will need to supplement it with multilingual models.
KoGPT earns its place as the pioneering open-source Korean LLM — it filled a genuine gap in 2021 and remains a cited baseline in peer-reviewed Korean NLP research. The 6B-parameter model delivers competitive zero-shot and few-shot performance on Korean benchmarks, and the Apache 2.0 code licence makes it straightforward to integrate into research pipelines. For non-commercial Korean language work, the value-per-dollar ratio is effectively infinite.
The caveats are significant, however. The CC-BY-NC-ND 4.0 weight licence blocks commercial deployment, no public inference API has ever been released, and the hardware floor (16–32GB GPU RAM) limits accessibility. The planned KoGPT 2.0 was cancelled in all but name, and Kakao itself now relies on OpenAI's GPT-5 inside KakaoTalk. Researchers working in Korean NLP should treat KoGPT as a solid foundational baseline, but evaluate newer alternatives — HyperCLOVA X Seed, EXAONE 3.5, or Upstage SOLAR — for production or cutting-edge research use.
What people say
KoGPT was the first large open Korean GPT model when Kakao Brain shipped it in November 2021 — 6.17 billion parameters, trained on Kakao's in-house 'ryan' corpus, and for a few years it was simply what you used if you needed Korean NLP and didn't want to build from scratch. It still holds up on the benchmarks that mattered then: 87.8% accuracy on NSMC sentiment classification, 78.0 F1 on YNAT topic tagging, both solid numbers for a five-year-old model.
The company that made it doesn't really exist anymore. Kakao folded Kakao Brain's AI team into Kakao Corp in mid-2024, then merged the remaining entity into subsidiary DK Techin that October, part of a broader push to cut overlapping units. A planned KoGPT 2.0 at up to 65B parameters never shipped. Instead, Kakao bet on a partnership with OpenAI, wiring ChatGPT directly into KakaoTalk, while building its own on-device assistant, Kanana, which is rolling out agent features and a dedicated Kanana Search through 2026.
None of that erases what KoGPT is good for today: free weights and Apache-licensed training code on GitHub and Hugging Face, still cited in Korean NLP papers as a baseline. The catch is the CC-BY-NC-ND 4.0 licence on the weights, which rules out commercial use outright, plus no public inference API and a 16-32GB GPU floor to run it locally. Treat it as a solid research artifact to benchmark against, not a foundation to build a product on — for that, look at HyperCLOVA X Seed, EXAONE 3.5, or Kakao's own Kanana line instead.
Summary of public user & expert reviews, compiled by RECATOOLS.
Notable facts
- KoGPT was trained on a corpus called the 'ryan' dataset — named internally at Kakao Brain — and uses a custom 64,512-token vocabulary optimised for Hangul morphology.
- The model requires a minimum 16GB GPU (float16) or 32GB GPU (float32) to run locally — making it one of the more hardware-demanding open research models of its era.
- Kakao founder Kim Beom-su personally served as the first CEO of Kakao Brain when it was spun out in February 2017 with 20 billion KRW (~$17.5M) in starting capital.
- Despite planning a KoGPT 2.0 with up to 65 billion parameters for Q3 2023, Kakao eventually pivoted to an OpenAI partnership — embedding GPT-5 into KakaoTalk, which hit 8 million users within three months of launch.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Kakao Brain (KoGPT)'s official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Kakao Brain (KoGPT) unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Kakao Brain (KoGPT) directly →
Spotted something out of date? Suggest an update →
Alternatives to Kakao Brain (KoGPT)
More in LLMs & Chat
