An anonymous AI model called Ox Alpha has emerged on OpenRouter, offering developers a free million-token context window and near unlimited usage for high-level reasoning and coding tasks. However, the unidentified provider retains all prompts and completions, leaving the tech industry guessing about its origin.
Ox Alpha quietly dropped onto OpenRouter as a stealth model from an anonymous third-party provider. OpenRouter classifies the software as a reasoning model engineered specifically for coding, sustained agentic work, and production workloads. It is built to handle long-horizon software engineering, complex logic, and heavy workflows combining text with visual context.
It is entirely free to use.
The open-source AI coding agent OpenCode noted on X that Ox Alpha would remain free for a week with near unlimited usage. According to OpenCode, the mysterious provider boasts a capacity of 100 trillion tokens per day. To put that in perspective, that throughput is roughly 100 times the volume of AI tokens that Visa processes in an entire month.
High-profile validation followed immediately. Stripe CEO Patrick Collison tested the model and posted on X that it is very impressive.
The Great Attribution Hunt: From Chinese Labs to Microsoft
When massive compute capacity appears without a corporate logo, the tech community starts forensic investigations. Early speculation pointed directly toward Chinese AI labs.
The online tech publication Wccftech suggested that Z.ai, the Chinese lab behind GLM-5, might be behind the release. Analysts noted that the company previously tested GLM-5 anonymously under the moniker Pony Alpha, while developers pointed out striking similarities in tokenizer behavior and model responses. This speculation arrives as Chinese labs like Zhipu, DeepSeek, and Moonshot AI aggressively challenge US competitors. Models from these firms frequently match leading American systems at a fraction of the cost, often under open-source frameworks. Moonshot’s Kimi K3, a 2.8 trillion-parameter open-weight model released in July, similarly upended Silicon Valley expectations for coding and agentic performance.
Yet the trail went cold almost as quickly as it heated up.
Wccftech later highlighted a competing technical analysis suggesting that Ox Alpha’s tokenizer could instead point toward Microsoft’s MAI family. By Saturday morning, AI analyst Andrew Curran posted on X that while GLM had been the leading theory on Friday night, by morning people seemed less sure of anything.
The Enterprise Data Dilemma
OpenRouter’s official listing explicitly states that prompts and completions are retained by the unidentified provider.
