Analysis: 49.3% of derivative language models posted to Hugging Face by Korean accounts this year are based on Chinese models, 46.8% on Qwen (community post)

Verified (fact)

According to an analysis posted on GeekNews, a Korean tech news community, 49.3% of the 1,845 language models newly released on Hugging Face by Korean accounts from January to August 2026 were built on Chinese models. The most-used base model was Alibaba's Qwen, with 864 models (46.8%), followed by Gemma (14.7%), Llama (13.8%), and EXAONE (9.3%). Models based on DeepSeek accounted for only 1.0%. Who wrote the original analysis and how the counts were compiled are not stated in the excerpt.

Why it matters (interpretation)

The figures show how heavily Korea's fine-tuning and derivative model ecosystem relies on Qwen. Regulatory variables around Chinese models, such as US measures excluding DeepSeek or China's proposed export controls on AI technology, could also affect Korean developers.

Sources

• Community 한국의 신규 AI 파생 모델 절반 가까이가 중국 Qwen 기반임 GeekNews

More about Alibaba

• 2026-10-22 P3 Dev tools Generally available Efficiency & infraLanguage models AWSAlibaba
  AWS outlines three ways to train Qwen3 on Trainium with torchtitan

  AWS laid out three ways to train Qwen3 with torchtitan on a single trn2.48xlarge instance. The paths use the Hugging Face Transformers...

• 2026-10-21 P2 Infrastructure Unconfirmed Efficiency & infra NVIDIAByteDanceAlibabaTencent
  "ByteDance, Alibaba, Tencent order over $35B in NVIDIA chips, about 3M RTX 5500 Pro units," per shared 'exclusive' (community, unconfirmed)

  A headline from an 'exclusive' article saying ByteDance, Alibaba, and Tencent ordered more than $35 billion worth of NVIDIA chips was...

• 2026-10-21 P2 Research Unconfirmed Efficiency & infraLanguage models EAGLEAlibaba
  Speculative decoding method EAGLE-4 released, "3.59x faster than vanilla decoding on Qwen3.8-27B" (community)

  News of the EAGLE-4 release was posted on Reddit's r/LocalLLaMA. In a shared Qwen3.8-27B benchmark, speedups over vanilla decoding were...

• 2026-10-21 P2 Open source Open weights Computer visionMultimodalData & evaluation XiaomiAlibaba
  Xiaomi releases 'Xiaomi-Lens-0', a 4B multimodal model that points to image regions at pixel level mid-answer, plus the 'LensBench' benchmark (report)

  According to Pandaily, Xiaomi's SeerRay-Lab open-sourced 'Xiaomi-Lens-0', a 4B multimodal model built on Qwen3.5-4B. While generating an...

• 2026-10-20 P2 Infrastructure Unconfirmed Efficiency & infra Alibaba
  Alibaba DAMO Academy says XuanTie RISC-V-based chip shipments top 6 billion, expanding into AI and autos (report)

  According to the South China Morning Post, Meng Jianyi, chief scientist at Alibaba's research arm DAMO Academy, disclosed the figure at...