Chinese AI Models Have Outpaced US Usage
Weekly token processing volume reached 62.22 trillion, extending a lead over U.S. competitors to 22 consecutive weeks.
Updated on Sept. 28, 2026 in Artificial Intelligence
Live Poll
Does higher AI model usage signify that a country is leading in technological innovation?
Chinese artificial intelligence models have maintained their market lead for the 22nd straight week, processing 62.22 trillion tokens compared to 14.2 trillion processed by U.S. models between September 21 and September 28, 2026. This period also saw three of the five most-used AI models on OpenRouter originate from Chinese developers.
Why it matters
The usage shift suggests a decisive move toward Chinese models for high-volume enterprise tasks, driven by competitive pricing and specialized performance. This trajectory indicates that developers are increasingly prioritizing price-performance ratios and specific workload compatibility over domestic platform preference.
DeepSeek V4.1 Flash led usage with 19.6 trillion tokens, followed by Z.ai GLM 5.3 Flash with 16.3 trillion tokens and the Tencent Hy4 preview with 9.64 trillion. The newly appeared Space Bunny Alpha, which features a 1 million token context window, processed nearly 14 trillion tokens in less than one week.
The players
DeepSeek
A Chinese AI research organization focused on high-performance large language models and efficient training architectures.
Tencent
A Chinese technology conglomerate with a deep stack in cloud computing, enterprise software, and consumer AI services.
MiniMax
A Chinese startup specializing in foundational large language models and autonomous agent technology.
Z.ai
A developer of specialized AI models known for high-throughput, low-latency performance in enterprise applications.
The details
Developers achieved this volume by shifting models from research-focused testing phases to full releases distributed across multiple platforms. These models are engineered for specific high-utility workloads, such as autonomous agent tasks—AI programs that execute series of actions to achieve a goal—and large-scale database searching. By emphasizing aggressive pricing and open-source availability, Chinese developers have accelerated the adoption of these tools for enterprise-grade applications.
Timeline
Sept 21, 2026 marked the beginning of the recorded weekly token usage period.
Sept 23, 2026 saw the introduction of the Space Bunny Alpha model on the OpenRouter platform.
Sept 28, 2026 concluded the latest measurement window for total token volume.
The Tech Race
This dominance marks a sustained divergence from the previous market trend where U.S. models maintained consistent usage leadership. The rapid adoption of new models like Space Bunny Alpha highlights a competitive research landscape where cycle times between model release and high-volume deployment are shrinking.
Enterprise users should anticipate faster integration of Chinese models into existing workflows due to their high performance in autonomous agent tasks and large context windows. Developers and organizations may need to evaluate these tools against current US-based alternatives to optimize costs for high-volume data operations.
The takeaway
The sustained volume gap suggests that price-performance and specialization are becoming the primary drivers of global model adoption. Watch for the official identification of the developer behind Space Bunny Alpha, which will confirm whether Chinese startups are successfully scaling new architectures faster than established global leaders.
Further reading
Explore deeper trends in model performance and adoption in our Artificial Intelligence section.
Source note: This article includes information reported by China Daily.
Live Poll
Does higher AI model usage signify that a country is leading in technological innovation?
