ETNews: 70.7% of the World's 300 Most Downloaded Language Models Are Chinese, and One Korean Model Is on the List
aillmopensourcenews

TL;DR
On November 18, 2026, ETNews, Korea's leading IT daily, reported VIDRAFT's analysis of the 300 language models that the world downloaded most on Hugging Face in the last 30 days. The 300 models account for 523.22 million downloads. Chinese models take 70.7% and US models 26.8%. One Korean model is on the list: VIDRAFT's POCKET-Darwin-180B.

What ETNews reported
| Point | What the article says |
|---|---|
| Scope | The 300 most downloaded language models on Hugging Face, last 30 days, 523.22 million downloads, as of November 17, 2026 |
| By country | China 70.7% (206 models), United States 26.8% (76), Europe 0.6% (4), and one Korean model |
| By family | Alibaba Qwen 60.3% with 173 models, GPT and OPT 10.5%, Google Gemma 10.5%, DeepSeek 4.9%, Zhipu GLM 2.9%, Meta Llama 2.2%, NVIDIA Nemotron 1.8% |
| Top model | GPT-2, released in 2019, is still first with 34.19 million downloads. Eight of the top 10 are Chinese families |
| Pace | 197 of the 300 models were released in 2026, and they take 56.7% of downloads |
| Format | Compressed formats such as GGUF, FP8 and NVFP4 take 49.1%. GGUF alone, the format that runs on laptops and PCs, takes 27.5% |
| Size | Models of 15 to 70 billion parameters take 32.7%. 33 models of 70 billion or more are on the list, with 9.4% of downloads |
| Korea | POCKET-Darwin-180B, 528 thousand downloads in 30 days, rank 260, the only Korean model in the top 300 |
The Korean model on the list
POCKET-Darwin-180B is a 180 billion parameter model compressed to run without a data center GPU, on a single PC with 128 GB of memory or on a laptop. It is one of the 33 large models in the top 300. The original model, Darwin-180B, is first among open-source models in SuperCLUE's November 2026 math and science reasoning evaluation.

Eleven Korean AI companies, 2026 releases only
The article also compares the language models that 11 major Korean AI companies released on Hugging Face since January 1, 2026. Nine of the 11 published 174 models this year.
| Company | Models released in 2026 | Downloads, last 30 days | Share |
|---|---|---|---|
| VIDRAFT | 99 | 1.12 million | 82.2% |
| SK Telecom | 5 | 86 thousand | |
| Krafton | 9 | 66 thousand | |
| LG AI Research | 11 | 50 thousand | |
| Kakao | 9 | 23 thousand |
By cumulative downloads this year, VIDRAFT has 3.86 million (58.0%), LG AI Research 1.97 million (29.6%) and SK Telecom 0.50 million (7.6%).

What the CEO said
Kim Minsik, CEO of VIDRAFT, said the AI that developers actually download is concentrated in a few families, and that more than half of the top ranks change to new models within a year. He said POCKET-Darwin-180B reaching the top 300 shows that developers choose even a very large model when it comes in a form they can run on their own machine.
How the count was made
The analysis uses the public Hugging Face API on November 17, 2026 and counts the last 30 days of downloads for language models. Country and family follow the organization that first developed the model. The same ranking and charts update every day on the Global LLM Download Leaderboard.
Links
- ETNews article (Korean): https://www.etnews.com/20261117000332
- Global LLM Download Leaderboard: https://huggingface.co/spaces/VIDraft/global-llm-leaderboard
- POCKET-Darwin-180B: https://huggingface.co/FINAL-Bench/POCKET-Darwin-180B-GGUF
- VIDRAFT press list: https://vidraft.net
FAQ
Which country's language models are downloaded most on Hugging Face?
In the 30 days to November 17, 2026, Chinese models took 70.7% of the downloads of the top 300 language models and US models took 26.8%.
Which model family is downloaded most?
Alibaba's Qwen family, with 173 models in the top 300 and 60.3% of downloads.
Is there a Korean model in the top 300?
Yes. VIDRAFT's POCKET-Darwin-180B is at rank 260 with 528 thousand downloads in 30 days.