Overview
- Ox Alpha appeared on OpenRouter as a labeled “stealth model” and was first reported by multiple outlets after it became publicly accessible in late August.
- OpenRouter’s listing says the model is built for coding and long‑horizon agentic tasks and can accept text, image and video with a roughly 1,000,000‑token context window.
- The model was offered free with near‑unlimited usage for about a week and an early post from OpenCode claimed the provider had capacity for roughly 100 trillion tokens per day.
- Developers and public figures quickly tested the model and some limited benchmarks and user reports said Ox Alpha outperformed top coding models, but those results are preliminary and not independently validated.
- Analysts are divided on origin—suggestions include Chinese labs such as Z.ai/GLM‑5 or Microsoft’s MAI family—while OpenRouter’s stealth preview conceals the upstream provider and leaves attribution and scale claims unresolved.