Hugging Face Report: Chinese Open Models Surpass US in Scale
RisingSayak · x · 2026-08-15
Hugging Face published the "State of Open Models: Summer 2026 Observations" report, analyzing ecosystem data from January to August 2026.
Key Findings:
- Ecosystem Growth: Public model repositories grew from 2.43M to 2.96M, datasets from 711k to 1M, and Spaces from 1.0M to 1.44M.
- Highly Skewed Distribution: About 85.6% of models have fewer than 200 lifetime downloads, while the top 1.5% of repositories account for 99.2% of all downloads.
- Shifting Frontier: Chinese labs have bypassed the traditional progression from small to large models. In most months of 2026, the largest open models from Chinese labs (754B–2.78T parameters) exceeded those from US labs (NVIDIA's Nemotron 3 Ultra at 561B was an exception).
- Vendor Strategies: Moonshot, MiniMax, Xiaomi, and Z.ai focus almost exclusively on models above 70B, posing challenges for local deployment, while Tencent and Alibaba cover the full range of sizes.
Related event: Hugging Face Report: Chinese Open-Source Models Now Outscale US(2 posts)→
More from Research
- Stanford researcher argues CoT monitoring has fragile foundations and long-term risks — maksym_andr · 2026-08-15
- Editor runs paper through ChatGPT after rejection; AI finds real errors missed by reviewer — skdh · 2026-08-15
- DeepSeek releases Harness with formal paper, inspired by Jane Street's functional programming — sudoraohacker · 2026-08-15
- Research: Rational AI adoption could destroy professional expertise — The Decoder · 2026-08-15
- New benchmark reveals poor visual perception in AI models; none reach 60% accuracy — The Decoder · 2026-08-15
- Why Claude.md Keeps Growing: The Cost of Deleting vs. Adding Instructions — 量子位 · 2026-08-15