A new anonymous model called Ox Alpha has appeared on OpenRouter as a large-context reasoning model whose provider has not been publicly named. OpenRouter’s listing says the model was released on August 20, 2026, and describes it as built for coding, sustained agentic work, production workloads, long-horizon software engineering, complex reasoning, and workflows that combine text with visual context. The most concrete specification is the context window. According to OpenRouter, Ox Alpha supports a 1,048,576-token context window and up to 131,072 completion tokens. The listing says the model accepts text, images, and video as input, and returns text. OpenRouter also makes clear what it is not claiming. The listing says Ox Alpha is developed and operated by a third-party provider that has chosen to remain anonymous during the preview. OpenRouter says it routes requests to the model but is not the model’s developer, owner, or provider. That routing setup is unusually narrow for OpenRouter: the listing says Ox Alpha is hosted by one provider and that OpenRouter forwards every request directly to it. OpenRouter’s page also says prompts and completions are retained by the provider and are not used for training, with other use governed by the Stealth Model Terms. For application developers, the API surface is presented as fitting existing OpenAI-style integrations. OpenRouter says its API is OpenAI-compatible and that most software development kits can work by swapping the base URL. The listing says Ox Alpha accepts tools and tool_choice for function calling, and supports response_format for JSON output, though without JSON-schema enforcement. Pricing is also part of the launch window. OpenRouter’s page says the pricing shown for Ox Alpha is zero, meaning users are not charged for prompt or completion tokens under the listed terms. The model’s actual economics are not otherwise explained in the provided material. Techmeme, summarizing a Wccftech report by Rohail Saleem, says Ox Alpha went viral after launching on OpenRouter and describes it as a stealth model from an unknown AI lab with 1M-token multimodal context and capacity for 100T tokens per day. The 100T tokens/day capacity figure appears in that Wccftech summary; it is not present in the OpenRouter excerpt provided here. The story is still developing because the model’s operator remains anonymous and several key claims depend on a single report or the model’s own listing. What is confirmed from OpenRouter is narrower but still notable: it lists a zero-priced-preview, long-context, multimodal reasoning model with an OpenAI-compatible API surface, while the identity and operating capacity behind it remain largely undisclosed. Who benefits: Developers building coding agents, long-context review tools, and multimodal workflows have a new model listing to evaluate under pricing that OpenRouter says is zero for prompt and completion tokens. OpenRouter is also the platform named in reports that say the model went viral after launching there. Who's exposed: Users face provider opacity: OpenRouter says prompts and completions are retained by the unnamed provider, even though they are not used for training.