China’s open‑source models lead token volume
New reporting shows Chinese AI models processed ~4.69 trillion tokens in a single week, putting China ahead of the US in open‑source model usage for the second week running. That surge is raising strategic concerns about where open‑model momentum and scale are consolidating globally. (investing.com)
OpenRouter’s usage dashboard and recent reporting place the 4.69 trillion‑token figure in the week of March 9–15, 2026, citing platform leaderboards and anonymized API aggregation data. (openrouter.ai) OpenRouter’s weekly window shows China’s token volume rose ~11.83% week‑on‑week while U.S. model token calls declined about 9.33% to roughly 3.294 trillion tokens in that same March 9–15 period. (binance.com) The platform’s leaderboards list Chinese labs in the top three slots by token volume and independent summaries put Chinese models at roughly a 61% share of OpenRouter token consumption in recent rankings. (openrouter.ai) Per‑model snapshots indicate MiniMax M2.5 has processed multiple trillions of tokens on OpenRouter (WorkOS reported ~2.45 trillion processed in a single week in February), and OpenRouter’s MiniMax model page shows a 196,608‑token context window and published pricing tiers. (workos.com) OpenRouter’s Kimi K2.5 listing shows a 262,144‑token context window and positions Kimi as a high‑throughput multimodal/agentic model, corroborated by the model’s Hugging Face and platform integration pages. (openrouter.ai) Coverage and platform commentary point to lower pricing and agent/coding‑heavy workloads as immediate drivers of the shift, and recent reporting that cites J.P. Morgan highlights forecasts of extremely large token demand growth in China toward 2030 (figures reported in coverage as ≈3,900 quadrillion tokens by 2030). (news.cgtn.com)
