Chinese open-weight models lead US in downloads and benchmark scores, analysis finds
Chinese AI labs have overtaken U.S. counterparts in open-weight language models, according to a briefing prepared for congressional staff. Since April 2025, Chinese models such as Alibaba’s Qwen family, Z.ai’s GLM‑5.3 series and Moonshot AI’s Kimi K3 have amassed about 1.6 billion of the 3.2 billion total Hugging Face downloads recorded by September 2026, roughly double the U.S. total. On the…
Key points
- Chinese open-weight models accounted for ~1.6 billion of 3.2 billion total Hugging Face downloads by Sep 2026, double the U.S. total.
- Top Chinese models (GLM‑5.3, GLM‑5.3‑Flash, Kimi K3) scored 42‑45 on AAII benchmark, versus 23‑26 for leading U.S. models.
- OpenRouter token processing rose from ~1 trillion weekly (Sep 2025) to ~80 trillion (Sep 2026), with Chinese models’ share climbing from ~70% to >80%.
OpenRouter token processing grew from roughly 1 trillion tokens per week in September 2025 to about 80 trillion in September 2026, with Chinese models’ market share rising from ~70% to over 80%. Academic papers on arXiv now mention Chinese open-weight models in over 40% of AI‑related submissions, compared with 30% for U.S. models. The author estimates that preventing distillation would only widen the gap by 1‑2 months, and notes growing regulatory uncertainty as open models enable new cybersecurity risks.
The analysis concludes that Chinese dominance in open-weight models reshapes AI research, commercial adoption, and geopolitical influence, urging continued U.S. investment in open models to mitigate risks and preserve competitiveness.
Model pages: GLM 5.3 → · Kimi K3 →
The story so far
6 episodes →- Chinese open-weight models lead US in downloads and benchmark scores, analysis findsthis story
The current balance of power in open models
Interconnects · 21 September 2026
Loading the full article…
This text was published by Interconnects and written by Nathan Lambert. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Hacker News discussion · 60 pointsnews.ycombinator.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- OpenAI offers U.S. university labs free ChatGPT Business seats with Rosalind access · 2 src
- Anthropic reports small AI models may self-replicate on IoT devices within months · 29 src
- Researchers propose LLM-referenced C-test for evaluating supra-intelligence · 8 src
- Researchers study permutation diffusion LLMs for recommendation reranking · 1 src
- Researchers compare Non-Generative verifier to Astra LLM for RAG claim checks · 1 src
Comments
via GitHub Discussions