Anonymous AI 'Ox Alpha' Enters Free Testing With Million Token Context
The scale of traffic flowing through an anonymous, unverified model suggests the organization behind Ox Alpha operates large-scale AI infrastructure, which is why the identity question matters as much as the model's benchmark results.
Reporting from 1 source: GIGAZINE.
AI platforms OpenRouter and OpenCode are running free trials of Ox Alpha, an anonymous AI model whose development organization and specifications are unknown. The model has a 1 million token context window, accepts text, image, and video inputs, and outputs text. OpenRouter reports daily traffic on the order of trillions of tokens. Ox Alpha was released on August 21, 2026.
Ox Alpha surfaced on OpenRouter as a stealth model, a deployment option that lets an organization test a model without revealing its official name or developer. The page lists a context window of 1 million tokens and multimodal input support covering text, images, and video, with text output.
The anonymous lab's infrastructure appears substantial. OpenRouter statistics show traffic on the order of trillions of tokens flowing daily. Nous Research, the developer of the Hermes Agent app that added Ox Alpha support, states the model can process 1000 trillion tokens per day. Ox Alpha is also free in OpenCode's cloud model service Go.
User tests are mixed. One user reported better results than Qwen3.8 27B and the model 4.5 on a GPU-based fluid simulation app. Another found Ox Alpha inferior to the model 5 on a three.js rocket engine 3D model, with longer output time. Cybersecurity benchmarks scored above GPT-5.6 Luna but below Grok 4.6 and GLM-5.3.
OpenRouter status pages show service has remained generally stable since launch, with a brief unstable period around 4:30 PM on August 24, 2026. The official name, developer, and release date remain unknown.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.