Image: SiliconAngle / siliconangle.com Nobody Knows Who Built Ox Alpha, the Free Frontier Coding Model Flooding OpenRouter
A mystery model called Ox Alpha has been running free on OpenRouter and OpenCode since August 20. It claims a 1M-token context window and 100 trillion tokens per day of capacity. The creator has not come forward, and fingerprinting points toward a company on the U.S. government's Entity List.
Since August 20, developers have had access to a frontier-class coding model on OpenRouter and OpenCode with no name, no known creator, and no explanation of what happens to their code once submitted. The model is called Ox Alpha. Free access ends sometime between today and August 27, depending on the platform.
The specs read well on paper. Ox Alpha has a 1,048,576-token context window, accepts text, images, and video (audio is rejected), and outputs up to 131,072 tokens. OpenCode described its capacity as “100 trillion tokens a day,” which the company framed as meaning developers would not bump into rate limits during the preview window.
Initial benchmarks looked better than they were
When Ox Alpha appeared, early testers were excited. Developer @davis7 shared results showing the model scoring above 80% on ten DeepSWE tasks, better than Claude Fable 5 at 65% and GPT-5.6-sol at 52%. That spread around quickly.
Full testing told a different story. When the benchmark expanded beyond that initial sample, Ox Alpha landed roughly at parity with GPT-5.6-sol, not ahead of it. The gap that generated the headlines did not survive scrutiny.
Who built it? Probably Z.ai, maybe Xiaomi, officially nobody
A developer using the handle unclecode built fingerprinting tools to compare Ox Alpha’s infrastructure signatures against known models. The result matched GLM-5.3, a model from Chinese AI company Z.ai (formerly Zhipu AI), on six of nine infrastructure probes. unclecode was careful to note the limits of the method: “matching fingerprints prove shared infrastructure, not identity.” Z.ai had previously released models anonymously before an official announcement, which fits the pattern here.
Xiaomi’s MiMo team has also come up as a possibility, though that speculation rests on weaker evidence.
No company has claimed Ox Alpha.
The Entity List problem
Z.ai’s predecessor, Zhipu AI, was added to the U.S. Commerce Department’s Entity List on January 16, 2025, for its role in advancing China’s military AI capabilities. Companies on the Entity List face restrictions on receiving U.S. technology exports.
That matters here because of how the data retention policies stack up. OpenRouter’s listing states it does not retain prompts or completions for training. But the broader anonymous preview terms that apply to Ox Alpha permit training and evaluation use. OpenCode promised zero retention on its side.
The problem is that developers routing code through an anonymous model cannot independently verify where it goes or what gets used for training. If the provider turns out to be Z.ai, that is a different risk calculation than using a model from a clearly identified vendor.
Coding tools collectively have already processed billions of tokens through Ox Alpha since launch.
Tight window, real tradeoff
The free access window is closing. OpenRouter users see access through roughly today; OpenCode extended it slightly to around August 27. After that, the model disappears or converts to paid access under whatever name the provider eventually chooses.
For developers evaluating it now: the benchmark claims are overstated, the creator is unknown, and the data handling policies are ambiguous. Those are not abstract concerns when you are pasting proprietary code into a model backed by anonymous infrastructure with a plausible connection to an Entity List company.
The model is genuinely capable, roughly in the GPT-5.6-sol range on coding tasks, which is useful. Whether that is worth the unknowns is a judgment call developers are making individually, because nobody has given them the information to make it well.
Sources: SiliconAngle, OpenRouter listing, Startup Fortune, ExplainX