Ox Alpha Appears on OpenRouter Free: Stealth Frontier Model Identified as GLM Architecture
A stealth model called Ox Alpha appeared on OpenRouter this week listed under the /stealth/ namespace — free to use, no training on user data, and carrying a 1M-token context window. OpenRouter’s own post described it as “a frontier model built for efficient coding, sustained agentic work, and real-world production use.”
Within hours of the listing going live, a community member published a GitHub gist identifying Ox Alpha as GLM-based — pointing to Zhipu AI, the Beijing lab behind GLM-5.3, which debuted on Arena’s Text Arena and Code Arena leaderboard on August 19.
What It Claims
OpenRouter positions Ox Alpha as purpose-built for three workloads:
- Long-horizon software engineering — sustained multi-step coding tasks, not just single-prompt completions
- Complex reasoning — problems requiring extended inference chains
- Multimodal production use — the model accepts text, image, and video input
The 1M-token context window puts it alongside GLM-5.3’s published specification. Pricing is zero, and the provider committed to not training on completions — a disclosure OpenRouter flagged explicitly, suggesting the operator anticipated the usual concerns about free frontier models.
The GLM Connection
GLM-5.3-max entered Arena on August 19, the same week Ox Alpha appeared. Zhipu AI has a track record of shipping capable frontier models quietly — GLM-5.2 arrived in June with MIT open weights and a 1M context window before making broader waves. If Ox Alpha is GLM-based, it follows a familiar pattern: deploy anonymously to gather production feedback, then surface the branding once the model has usage data.
The stealth namespace on OpenRouter has been used for this before. Ant Group’s Ling-2.6 arrived under a stealth handle in April; Elephant Alpha, OpenRouter’s own 100B model, launched free in a similar fashion.
Stakes for Coding Agents
Free frontier-class models at 1M-token context change the arithmetic for teams building coding agents. Most competitive coding workloads today run on Claude Sonnet 4.6 or GPT-5.4 at measurable per-token cost. An unmetered GLM-class model with equivalent or better agentic positioning removes cost as a barrier to experimentation.
Whether Ox Alpha benchmarks at the level of its positioning is not yet established. No SWE-bench or Terminal-Bench result has been published. The model is live on OpenRouter now.