Chinese AI models lead global token usage for 22nd week; anonymous model enters top three
Chinese AI large language models have maintained the top global position in weekly token usage for 22 consecutive weeks, according to OpenRouter data for September 21-27. Global total call volume reached 146 trillion tokens, up 13.18% week-over-week. Chinese models accounted for 62.22 trillion tokens, while US models had 14.2 trillion tokens. An anonymous model named Space Bunny (or Space Bunny Alpha) surged to third place with 13.9 trillion tokens. Tokenizer tests suggest it may originate from Chinese AI company MiniMax, though MiniMax has not confirmed this. MiniMax released its M3.1-Flash-Preview model on September 28.
Reference imageEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Common ground
- China's 22-week lead in AI call volume and cost efficiency (models priced 1/20 to 1/60 of Western alternatives) represents a genuine structural shift in global tech power.
- The anonymous 'Space Bunny' model shows that innovation can thrive outside Silicon Valley's hype-driven system.
- Western media's framing of 'China catching up' is outdated and ignores the data showing China's current lead.
- The Global South benefits from affordable AI options, reducing dependency on expensive Western models.
Points of contention
- Eastern agents see China's state-backed industrial policy as a strength, while the regional agent argues it raises concerns about labor conditions and regulatory uncertainty.
- Eastern agents dismiss critiques of labor and data consent as selective moralizing, but the regional agent insists these issues are universal and must be addressed for all poles.
- The regional agent views the 'Space Bunny' anonymity as a sign of regulatory hedging, while Eastern agents see it as a sign of rapid, unconstrained innovation.
Blind spots
- The debate overlooks the long-term stability of Chinese AI models, especially whether anonymous or state-tied models could disappear due to political shifts.
- There is little discussion of how the Global South's dependency on cheap Chinese AI might create new forms of technological hegemony, not just liberation.
- The human cost of AI development—like labor conditions in data centers—is acknowledged but not deeply explored with concrete evidence from all sides.
WorldAttention’s read
This debate shows that China's 22-week AI lead and cost advantages are real and reshaping global tech, but the conversation is too focused on a victory narrative. Both sides agree the multipolar shift is happening, but they clash over whether China's model is a clean alternative or just a different set of problems. The blind spots are critical: we need to ask if cheap AI from China will stay reliable, if the Global South is trading one dependency for another, and how to ensure accountability for labor and data practices everywhere. The real test isn't who leads in token counts—it's whether these tools truly serve people with transparency and stability, not just affordability.
Reporting timeline
China's AI Model Call Volume Leads Globally; Anonymous 'Space Bunny' Linked to MiniMax
China's AI large model weekly call volume has surpassed the United States for 22 consecutive weeks, according to OpenRouter data from September 21-27. Global total call volume reached 146 trillion tokens, up 13.18% week-over-week. China's weekly call volume was 62.22 trillion tokens, down 7.77%, while the US had 14.2 trillion tokens, down 0.07%. Among the top five models by call volume, three were Chinese: DeepSeek V4.1 Flash (first, 19.6 trillion tokens), GLM 5.3 Flash (second, 16.3 trillion), and Tencent Hy4 preview (fourth, 9.64 trillion). An anonymous model named Space Bunny (also called 'Yutu') ranked third with 13.9 trillion tokens. Developers have linked Space Bunny to MiniMax's M3.1-Flash-Preview via tokenizer tests, though MiniMax has not confirmed. MiniMax CEO Yan Junjie stated in an earnings call that text model unit compute throughput has tripled in two months, with the goal of reducing M3.1 inference costs to one-third of M3's launch cost. He expects text business to drive gross margin improvement in the second half of the year and into 2025, and that billing models may shift from per-token to per-task or value-based pricing.
China AI Models Lead Global Calls for 22nd Week; Anonymous 'Space Rabbit' Enters Top Three
According to the latest data from OpenRouter, global AI large model total calls reached 146 trillion tokens in the week of September 21-27, a 13.18% increase from the previous week. Among the listed AI models, Chinese AI large models recorded 62.22 trillion tokens in weekly calls, a 7.77% decline from the prior week, while US AI large models had 14.2 trillion tokens, a 0.07% decline. China's large model weekly calls have surpassed the US for 22 consecutive weeks, maintaining the global top position. An anonymous model named 'Space Rabbit' surged into the top three, with tests suggesting it may originate from Chinese AI company MiniMax.
Read sourceChina AI Model Call Volume Leads US for 22nd Consecutive Week, OpenRouter Data Shows
According to data from OpenRouter, as reported by Jin10 on September 28, global AI large model call volume reached 146 trillion tokens in the week of September 21-27, a 13.18% increase from the previous week. Chinese AI large models accounted for 62.22 trillion tokens of that total, a 7.77% decline week-over-week, while US AI large models accounted for 14.2 trillion tokens, a 0.07% decline. Despite the weekly drop, China's AI large model call volume has surpassed that of the United States for 22 consecutive weeks, maintaining the top global position. The data was sourced from OpenRouter and reported by the Daily Economic News.
Read sourceShow 3 older updatesHide older updates
China's AI Model Call Volume Leads for 22nd Week; Anonymous 'Space Bunny' Enters Top Three
According to OpenRouter data analyzed by National Business Daily, China's AI large language models have led global weekly call volume for 22 consecutive weeks as of September 21-27. Total global AI model call volume reached 146 trillion tokens, up 13.18% week-over-week. China's models accounted for 62.22 trillion tokens (down 7.77%), while US models had 14.2 trillion tokens (down 0.07%). In the global top five, three Chinese models ranked: DeepSeek V4.1 Flash (first, 19.6 trillion tokens, up 24%), Zhipu GLM 5.3 Flash (second, 16.3 trillion tokens, up 16%), and Tencent Hy4 preview (fourth, 9.64 trillion tokens, down 23%). An anonymous model, Space Bunny Alpha, surged to third place with 13.9 trillion tokens after launching on Openrouter on September 23. Tokenizer fingerprint tests suggest it may come from MiniMax, though this is unconfirmed. Xiaomi's MiMo-V2.6-Flash entered at eighth place with 5.51 trillion tokens, following the open-source release of the MiMo-V2.6 series on September 22. Xiaomi claims the Pro variant scores 46 on the AA Intelligence Index, surpassing Kimi K3 and Qwen3.8 Max, but still trails top closed-source models like Claude Fable 5.1 and GPT-6 Astra. The MiMo-V2.6 series is priced at 1/20 to 1/60 of comparable overseas models. Previous entrants Xiaomi MiMo-V2.5 and Tencent Hy3 dropped out of the rankings.
Read sourceChina's AI Model Call Volume Leads for 22nd Week; Anonymous 'Space Bunny' Enters Top Three
According to an analysis by National Business Daily based on OpenRouter data, global AI large model call volume reached 146 trillion tokens in the week of September 21-27, up 13.18% week-on-week. Chinese AI models accounted for 62.22 trillion tokens, a 7.77% decline, while US models had 14.2 trillion tokens, down 0.07%. China has led the US for 22 consecutive weeks. Among the top five models by call volume, three are Chinese: DeepSeek V4.1 Flash (first, 19.6 trillion tokens), GLM 5.3 Flash (second, 16.3 trillion tokens), and anonymous model Space Bunny Alpha (third, 13.9 trillion tokens). Space Bunny Alpha, launched on OpenRouter on September 23, features fast reasoning and programming with a 1 million token context window. A tokenizer fingerprint test suggests it may come from MiniMax, though this is unconfirmed. Xiaomi's MiMo-V2.6-Flash ranked eighth with 5.51 trillion tokens. Xiaomi claims its MiMo-V2.6-Pro is the strongest open-source model, scoring 46 on the AA Index, but still trails top closed-source models. Its pricing is 1/20 to 1/60 of overseas models at similar intelligence levels.
Read sourceChina's AI Model Usage Leads Globally; Anonymous 'Space Bunny' Model Linked to MiniMax
According to a report by The Paper cited by East Money, China's large language models (LLMs) have maintained the top global position in weekly token usage for 22 consecutive weeks. Data from the model aggregation platform OpenRouter shows that from September 21 to 27, global AI model total calls reached 146 trillion tokens, a 13.18% increase week-over-week. Chinese AI models accounted for 62.22 trillion tokens, while US models accounted for 14.2 trillion tokens. Among the top five models by usage, DeepSeek V4.1 Flash ranked first, followed by Zhipu GLM 5.3 Flash, and Tencent Hy4 preview. An anonymous model named 'Space Bunny' (also called 'Yutu') surged to third place with 13.9 trillion tokens in weekly usage. Developers have speculated, based on tokenizer tests, that Space Bunny may be a preview version of MiniMax's M3.1 Flash model, though MiniMax has not confirmed this. Separately, MiniMax released its M3.1-Flash-Preview text model on September 28, supporting native multimodality and a million-token context window. MiniMax CEO Yan Junjie stated in an earnings call that the company aims to reduce M3.1 inference costs to about one-third of the original M3 model, with some savings passed to customers to expand token scale and some retained to improve gross margins. He added that multimodal and voice services currently have higher gross margins, and text model revenue is growing rapidly while inference costs decline, making text a key driver for overall margin improvement.
Read source