Beijing-based AI lab Z.AI, widely known as Zhipu, has confirmed that the mystery stealth model "Ox Alpha" is an upcoming release in its GLM model series. The company confirmed the model's identity in response to inquiries from Bloomberg on Wednesday and stated it will release the open model weights tonight.
Ox Alpha first appeared on August 20, 2026, as an uncredited stealth model on the OpenRouter model routing platform. Operating under a free evaluation tier, the model quickly surged to the top of developer usage leaderboards due to its combination of high inference throughput, strong agentic coding capabilities, and zero API cost during the preview phase.
Model Specifications and Developer Findings
According to OpenRouter's model specifications, Ox Alpha features a native context window of 1,048,576 tokens (1M tokens) and supports multimodal inputs across text, images, and video. The model was engineered specifically for sustained software development workflows, long-horizon agent execution, and structured reasoning tasks.
In the days following its anonymous release, developers and independent researchers analyzed the model's output syntax and tokenization behavior. Comparative tests on coding benchmarks and markdown table structuring revealed strong stylistic overlap with Zhipu's GLM-5 architecture, leading community members to speculate that Ox Alpha was a preview checkpoint of GLM-5.3 or GLM-5.3-Flash.
Zhipu previously disclosed that the GLM-5.3 iteration builds upon the same 700-billion-parameter base model foundation used in GLM-5.2, with primary architectural gains derived from targeted post-training for code generation, agentic tool invocation, and long-context coherence.
Strategic Context in the Open-Weight Ecosystem
The official confirmation and imminent open-weight release place Zhipu in direct competition with frontier open models from DeepSeek, Moonshot AI, and Alibaba. By open-sourcing the weights for a model that has already demonstrated competitive performance against proprietary commercial APIs on developer platforms, Zhipu reinforces the accelerating trend of frontier-class capabilities moving to open deployment.
The release also highlights how AI laboratories are increasingly utilizing anonymous stealth rollouts on developer hubs like OpenRouter to stress-test inference infrastructure, gather telemetry, and benchmark post-training optimizations prior to public weight distribution.
