Chinese models dominate token use
Chinese models like DeepSeek and MiniMax have surged to the top of global token consumption, overtaking U.S. rivals thanks to rapid open‑source releases and domestic adoption. That shift means more permissively licensed, high‑usage models are available — but also raises long‑term support and compliance questions for teams integrating non‑U.S. stacks. (ft.com)
OpenRouter’s weekly leaderboard published February 24, 2026 showed the platform’s top 10 models consumed roughly 8.7 trillion tokens, with models developed in China accounting for about 5.3 trillion tokens — roughly 61% of that total. (dataconomy.com) MiniMax’s M2.5 was the single biggest driver on that leaderboard, processing about 2.45 trillion tokens in a single week and rising to the #1 spot on OpenRouter’s ranking. (dataconomy.com) OpenRouter’s model pages list MiniMax M2 variants at roughly $0.255 per million input tokens and $1 per million output tokens, pricing that multiple outlets and developer threads have estimated to be on the order of 17–20× cheaper than some leading Western proprietary models. (openrouter.ai) Platform and industry coverage point to agent-style, multi-step workflows and programming/coding tool use as the main short-term token-growth engines, with reports linking the surge to adoption of agent frameworks and coding agents such as OpenClaw. (pandaily.com) A high-profile stealth release illustrated the shift: an anonymous “Hunter Alpha” model that surfaced on OpenRouter on March 11 was publicly attributed to Xiaomi on March 18 and described as an internal MiMo‑V2‑Pro test build with about 1 trillion parameters and a reported 1‑million‑token context window. (theoutpost.ai) Coverage across industry analysts and platforms notes that the same openness and low pricing that drove developer adoption are also prompting enterprise caution — procurement, export‑control and chip‑supply constraints, plus data‑residency and compliance questions, are repeatedly cited as limits on replacing U.S. stacks in core enterprise systems. (thechinaacademy.org)
