A “stealth” model named Ox Alpha appeared on OpenRouter and OpenCode yesterday. It was free for users for a week, and the platforms were offering 100 trillion daily tokens for its use. Many users felt that the model seemed to be at par with GPT 5.6 Sol and Fable 5 in performance, with one user saying that it beat them both on the Deep SWE benchmark.
There has been plenty of speculation on which lab is behind the OX Alpha model. People have been wondering if it’s a breakthrough from a Chinese lab, or whether it’s from the US. These are the most popular theories around the Ox Alpha model on X.
1. Z.AI #
Several commentators have said that Ox Alpha is from the Z.AI stable. They point out that the tokenizer used in the model — which decides how sub-words are processed internally — corresponds to the Z.AI family of models. Jailbreaker Pliny the Liberator also came to this conclusion with his analysis.
There are some, however, who point out to the generous free usage that Ox Alpha is giving out which adds up to hundreds of trillions of tokens per day. Some say that it’s unlikely that a Chinese lab has this much compute to test a model with, and that makes it unlikely that the model is from the GLM series.
2. Microsoft #
But some other X users have interpreted the tokeniser results differently. Robert Lukoszko, CEO of Stormy, said that the model was likely from Microsoft. “Based on tokenization, stealth/ox–** alpha** is a cl100k_base-tokenizer model — which rules out OpenAI, Google, Anthropic, xAI and every Chinese frontier lab and matches only Microsoft’s Phi/MAI lineage (with IBM Granite as the outside bet) — so it is most likely Microsoft’s unreleased MAI frontier model, i.e. “MAI‑2″,” he posted on X. Microsoft has recently come out with a capable image model, so it’s possible that it has created a frontier-level general model as well. Also, Microsoft is likely one of the few companies that can afford to give out so much free compute to test the model.
3. Google #
Google has just come out with Gemini 3.7 Flash which appears to have exceeded most expectations for a Flash model. Google’s Gemini 3.5 has long been delayed, so it’s possible that Ox Alpha could be Gemini 3.5 Pro, or even Gemini 3.8 Flash. Interestingly, some Google employees have been ‘vaugeposting’ about a strong release, which could mean that Ox Alpha is Gemini 3.5 Pro. Like Microsoft, Google is a lab that could possibly have enough spare compute to spare to test out a model this way.
However, Ox Alpha appears to have failed a popular vision test that Gemini models pass, which could indicate that it is not from the Gemini series.
4. Composer #
Some have said that Ox Alpha could be Cursor’s new Composer model. xAI is another lab that could power all this token usage, and could be behind the new release.
5. Bytedance #
Another theory says that that Bytedance could be behind the model. This theory states that while the model appears to be from China, Z.AI doesn’t have the compute to give away all these tokens for free. Bytedance, on the other hand, could possibly have the compute to test out a model in this way.
The whole thing a bit of a mystery, and one that X users have sunk their teeth into. But regardless of which lab ends up being behind Ox Alpha, the strategy has created plenty of buzz, and other labs could soon begin marketing their models through stealth launches like this one.