On the world's largest neutral model router, Chinese-built models now command a share of usage that was unthinkable a year ago. This is the chart that should worry Western labs.
ort.fabryka.ai · August 05, 2026
69%
Chinese-origin share of tokens, today
31%
Western share of tokens, today
2026-03-21
when the lines crossed
A year ago, using a Chinese model in production was a curiosity. Today it is the default for a large slice of OpenRouter's traffic. DeepSeek, Qwen, MiniMax, Moonshot, Z.ai, Xiaomi and Tencent — collectively, Chinese labs have gone from rounding error to plurality.
Share of OpenRouter tokens — Chinese-origin vs Western models (%, 7-day avg)
Dashed marker = Hermes Agent launch (Feb 2026) — the agent app that defaults to cheap Chinese models, the candidate trigger for the China-share jump. Date is approximate (public announcement; OpenRouter exposes no app launch date).
Chinese-origin models first overtook Western ones around 2026-03-21 and haven't looked back. The mechanism is not mystery — it is price. Open-weight Chinese models deliver
good-enough quality at a fraction of frontier pricing, and developers route accordingly. Volume
follows cost long before reputation catches up.
The West still wins the revenue line. It is losing the volume line — and volume is the leading indicator.
The two clocks — share of volume (solid) vs revenue at list prices (dashed), Chinese vs Western
Revenue = tokens × each model's list price (prompt+completion, real usage mix). Revenue/price lines start 2026-04-22 — before that, too many models OpenRouter has since delisted are unpriced (no historical price series exists yet), so we don't fabricate it. Volume is full history. Volume has crossed to China (69% vs 31%); revenue is still Western (66% vs 34%).
The gate: why revenue hasn't followed volume
China already won volume — but a Chinese token is worth far less than a Western one. Today the
West lists at $2.52/M tokens against China's $0.38/M: a Western token is
~6.6× more valuable. That multiple is the gate. At today's price gap, China would need
~87% of all volume — not the 69% it holds now — to reach half of revenue.
So the revenue line can sit still even as the volume line runs away.
34%
Chinese revenue share, today
~6.6×
Western $/token premium
~87%
volume China needs for 50% of revenue
The signal that actually matters — blended price per million tokens, China vs Western ($, 7-day avg). Revenue share crosses only as these two lines converge.
The race to watch is no longer volume share — it's China's price per token. Revenue moves only when premium Chinese models (DeepSeek V4 Pro, GLM, Qwen Max) lift that line toward the West's.
Who is doing the taking — and the giving
Chinese gainers (share)
Deepseek+19.7pp
Xiaomi+11.6pp
Tencent+11.1pp
Z-Ai+4.0pp
Stepfun+1.9pp
Western losers (share)
Google-12.3pp
Anthropic-8.0pp
X-Ai-5.2pp
Openai-4.6pp
Meta-Llama-1.1pp
The handoff is close to one-for-one: the share Western incumbents shed is the share Chinese
challengers absorb. Inside multi-model agent apps, swapping a Western default for a Chinese one
is a config change — and at these price gaps, a tempting one.
Why it matters
Cost is eating quality's lunch at the bottom. The high-value slice still pays for Western frontier models; everything price-sensitive is migrating east.
Volume is a beachhead. Chinese labs are building the developer habits and tooling defaults that revenue eventually follows.
The neutral router makes switching frictionless. OpenRouter abstracts the vendor away — brand loyalty erodes when the next model is one string away.