When an unknown AI lab drops an anonymous model - Ox Alpha - for free, while declaring that they have the capacity to process up to 100 trillion tokens per day, we can reasonably surmise that the 'Wild West' age of cutthroat competition for AI models has arrived.
Given the fact that not many AI labs have the capacity to process the declared quantum of tokens per day, the internet is going bonkers trying to figure out if the Ox Alpha is an unreleased model in the GLM family, DeepSeek's V4-Flash, or SpaceXAI's unreleased Grok model.
As mentioned earlier, someone anonymously dropped Ox Alpha on OpenRouter and OpenCode on August 20, offering a 1-million-token multi-modal (text, audio, video) context window, a 131,072-token output limit, and a capacity to process 100 trillion tokens per day for a week entirely free of cost.
For the benefit of those who might not be aware, the Ox Alpha is a reasoning model "designed for coding, sustained agentic work, and production workloads — long-horizon software engineering, complex reasoning, and workflows that combine text with visual context," as per its own claims.
Coming back, these extremely generous limits have obviously set the proverbial tongues wagging across the entire breadth of the AI sphere. The strongest evidence points to Zhipu as the AI lab behind Ox Alpha, especially as it previously live-tested its GLM-5 model under the Pony Alpha moniker . There is a plethora of anecdotal evidence as well that supports this hypothesis:
- The Ox Alpha's video encoding pipelines, tokenizer math, and character-for-character responses closely resemble Zhipu's architecture.
- On third-party leaderboards such as Kingbench, Ox Alpha clinched a score of 87.5 percent , sitting just behind the GLM-5.3's score of 91.25 percent.
Given the fact that GLM-5.3, lacks native vision or video capabilities on public endpoints, while Ox Alpha actively promotes text, image, and video inputs, we can reasonably surmise that it's an unreleased, heavily upgraded GLM-class model, if Zhipu is indeed behind this effort.
Finally, Zhipu does have the compute resources to offer 100 trillion tokens per day for free. After all, it already runs multiple 10,000-GPU clusters and has just energized a 1GW data center.
Even so, there are others who point to Ox Alpha's uber-aggressive approach here as being synonymous with DeepSeek's playbook. What's more, DeepSeek's V4 model was initially launched as Hunter Alpha on March 11.
DeepSeek, however, does not appear to have enough compute resources to sustain Ox Alpha's throughput, as indicated by its recent decision to raise the price of the V4-Flash model . Also, DeepSeek V4 Flash does not natively scale to or favor the specific 1,048,576 context window behavior seen here, and its tokenization math deviates sharply from Ox Alpha's outputs.
As for SpaceXAI, it certainly has the compute to sustain Ox Alpha's throughput. But then, Ox Alpha's policy guidelines and nuanced boundaries regarding highly sensitive Chinese political topics suggest a model trained on mainland Chinese data sets. In the meantime, the mystery continues to swirl.
Follow Wccftech on Google to get more of our news coverage in your feeds.