The most-used model on OpenRouter right now belongs to nobody anyone can name. In the routing platform's weekly leaderboard, which measures tokens processed through its API, the top slot for the week ending October 4, 2026 belongs to a model called Space Bunny Alpha — published by an author listed simply as "stealth," with no company name, no model card and no announcement. It moved 38.7 trillion tokens in seven days, a 179% week-over-week jump, roughly one and a half times the traffic of the runner-up.
That runner-up is not a Western frontier lab either. DeepSeek's V4.1 Flash sits at number two with 25.6 trillion tokens, up 31%. The rest of the top five is an all-Chinese open-weight block: Zhipu's GLM 5.3 Flash at number three (9.73 trillion tokens, down 40% from a spike the week before), Xiaomi's MiMo-V2.6-Flash at number four (9.63 trillion, up 75%), and Tencent's Hy4 preview at number five (6.51 trillion). OpenAI's GPT-6 Luna, the highest-ranked Western model, comes in sixth at 6.18 trillion tokens — up a brisk 116%, but from a smaller base.
The picture is just as stark when traffic is measured by requests rather than tokens. For the week beginning September 28, DeepSeek led all model authors with 21.8% of text requests on the platform, narrowly ahead of Google at 21.5% and OpenAI at 18.2%. Authors grouped as "stealth" — a category that now includes Space Bunny Alpha — collectively took 10.5% of requests, a 296% increase over the prior week. On the trailing-thirty-day view, DeepSeek V4.1 Flash has processed 65.9 trillion tokens since first appearing, with GLM 5.3 Flash at 54.9 trillion, up 252%.
Anonymous "stealth" listings are a known quirk of OpenRouter: labs and well-funded groups sometimes ship models under codenames to gather real-world traffic data before an official launch. The pattern has preceded several high-profile releases in the past, and the +296% surge in the stealth category's request share suggests at least one undisclosed model is being aggressively tested in production right now. Whether Space Bunny Alpha is a Chinese lab, an American one, or something else entirely cannot be determined from the listing alone.
Two caveats matter when reading these numbers. First, OpenRouter's rankings count tokens processed, not users or spend — token totals reward verbose models and high-volume automation, so they measure usage, not capability. On the platform's benchmark tab, which uses the Artificial Analysis Intelligence Index, the top of the leaderboard still belongs to Anthropic's Claude Opus 5.5 (57.6) and Sonnet 5.5 (56.0), with Qwen3.8 Max and GPT-6 Astra close behind. Second, OpenRouter traffic is a sample of the aggregator market, not the whole industry; direct API usage at the big labs never shows up here.
Still, the trend line is hard to dismiss. Fresh entrants are climbing fast on both sides: Anthropic's Claude Sonnet 5.5 and OpenAI's GPT-6.1 Sol debuted near the top of the weekly trending chart with roughly 790 billion and 731 billion tokens respectively in their first days. But the volume story of this quarter is open-weight Chinese models — cheap, fast Flash-tier variants that coding agents and automation pipelines burn by the trillion. Four of the five most-used models on the largest open router now come from Chinese labs, and the one model outrunning them all refuses to say whose it is.
Comments (0)
Log in to join the discussion
Log InNo comments yet