Ox Alpha mystery model offers free million-token context on OpenRouter

An anonymous AI model called Ox Alpha has appeared on OpenRouter and OpenCode with unusually high free usage limits, including a one-million-token multimodal context window. Rohail Saleem reports for Wccftech that the service claims it can process up to 100 trillion tokens per day for one week without charge.

Ox Alpha is presented as a reasoning model for software development, extended AI agent tasks, and production workflows. Its listing says it can accept text, images, and video, while generating responses of up to 131,072 tokens. A context window of this size could allow users to submit very large codebases, lengthy documents, or extensive project materials in a single request.

Competing theories over the model’s origin

The identity of the lab behind the release remains unconfirmed. Saleem writes that early technical comparisons pointed to Chinese AI company Zhipu and a possible unreleased model in its GLM family. Observers cited similarities in video processing, tokenization behaviour, and model responses. Ox Alpha also reportedly showed policy boundaries around politically sensitive Chinese topics.

Other theories have included DeepSeek and xAI. However, the report says critics question whether DeepSeek has sufficient computing capacity to sustain the stated volume. The model’s context-window behaviour and tokenization also appear different from DeepSeek’s publicly known systems.

A later update raises a different possibility: an unreleased Microsoft MAI model. According to the update, Ox Alpha appears to use the cl100k_base tokenizer, a token-splitting system associated with OpenAI-derived tooling and used in Microsoft’s Phi and MAI model lineage. That finding would make a Chinese origin less likely, though it does not establish ownership.

Neither Microsoft, Zhipu, DeepSeek nor xAI has publicly confirmed a connection to Ox Alpha. For users, the release offers access to substantial capacity, but its anonymous operator and short free-access period make independent testing and careful handling of sensitive data essential.