Ox Alpha: a free 1M-context frontier coding model on OpenRouter
Anonymous and free for now — also live in OpenCode — with image and video input. Tokenizer sleuths suspect GLM-5.3, though your prompts are retained.
Copy markdown
A free frontier model appeared from nowhere
OpenRouter listed 'Ox Alpha' (model id stealth/ox-alpha) at $0 in and $0 out — a reasoning model built for coding and long-horizon agent work, with a 1,048,576-token context, 131K max output, and text, image and video input. It speaks the OpenAI-compatible API, so you point your existing client at it and go.
Free for a week inside OpenCode too
Beyond the raw API, Ox Alpha is wired into OpenCode free for a week with 'near-unlimited' usage and zero data retention — OpenCode claims headroom for 100 trillion tokens a day. Reality check: some users got rate-limited on their very first prompt, so treat that ceiling as aspirational.
Who's behind it? Nobody's saying
OpenRouter won't name the provider. Developer 'dax' reports the tokenizer lines up almost exactly with Z.ai's GLM-5.3; others point to Xiaomi's MiMo team, whose CFO recently teased a new model. Reportedly one of the two — but don't wire compliance or fallbacks around a guessed identity.
The catch: your prompts are logged
For the preview window the anonymous provider retains prompts and completions (it says it won't train on them), no independent benchmarks exist yet, and free capacity can vanish without notice. Perfect for burning tokens on throwaway agent runs and whole-repo experiments this weekend — not for anything carrying production secrets.