Ecosystem signals

Capability is one thing; adoption is another. Two lenses on demand — where tokens get routed in production (OpenRouter) and which model weights the world is pulling down (Hugging Face). Verified 2026-07-09.

OpenRouter — routed usage

69.1%

of routed tokens go to open weights

Open passed closed around Q1 2026 and kept climbing — up from ~22% a year ago.

16.3%

of all tokens route through DeepSeek

The largest single provider on the router — ahead of Google, OpenAI, and Anthropic individually.

6 of 10

top models by volume are Chinese open weights

DeepSeek V4 Flash is #1, Tencent's Hy3 Preview #2 (listed at $0.063 per 1M input tokens).

~12%

Anthropic token share — but the revenue lead

Premium per-token pricing means Claude captures the most dollars while open models take the volume.

Open vs closed token share

Share of routed tokens. Endpoints reported; interior points approximate (≈).

Closed / proprietary
Open weights & source
Routed tokens are the strictest adoption test — real workloads, paid per token. Open's climb from ~22% to 69.1% in a year happened without holding a single #1 leaderboard slot; price did the work.

Volume is not value

Why both sides claim they're winning.

Token share measures volume, not value. Cheap open models absorb bulk workloads (extraction, summarization, high-volume agents) while premium closed models keep the tasks where a bad output costs real money. Both statements are true at once: open weights run most tokens; closed models earn most revenue.

The tell in the data

US labs' combined token share fell from ~70% to ~30% in twelve months, yet Anthropic — at ~12% of tokens — still captures the most revenue on the router. Cheap tokens flow to open weights; expensive decisions stay closed.

Hugging Face — open-model downloads

41%

of past-year Hub downloads are Chinese models

Over 45% on recently uploaded models — Hugging Face reports China overtook the US on recent adoption.

1B+

cumulative Qwen downloads by March 2026

The fastest any open family has reached the milestone; 153.6M downloads in February alone — more than the next eight combined.

113k+

derivative models built on Qwen

More derivatives than Google and Meta combined — the ecosystem moat compounding on top of raw downloads.

5 of 10

trending models from Chinese labs

Highest concentration on record after the June frontier wave; DeepSeek's newest Flash took #1 within a week.

Download share by lab origin

Share of open-model downloads by publishing lab's region. Endpoints from Hugging Face reporting; interior points approximate (≈).

China
US
Europe
Other
Downloads measure who developers build on, not who ships the best demo. On recently uploaded models — the leading indicator — Chinese labs are already past 45%.

Trending on the Hub right now

After the June frontier wave. Download counts move daily, so ranks are shown without stale precision.

1

DeepSeek V4.1 Flash

#1 within a week of release

DeepSeek
China
2

Qwen 3.7

1B+ cumulative family downloads

Alibaba
China
3

GLM-5.2

MIT license; open frontier leader

Z.ai
China
4

Llama 4.5

Meta
US
5

Gemma 4

Google
US

5 of the top 10 trending models are from Chinese labs — the highest concentration on record.