Ox Alpha: free 1M context, from a provider who won't sign it
Ox Alpha is a free million-token coding model on OpenRouter from an anonymous provider that retains your prompts. Read that sentence twice before you route an agent at it.
A model called Ox Alpha showed up on OpenRouter last week with a million-token context window, 131,072 tokens of output, and a price of zero. Developers started pointing coding agents at it immediately. The listing also says the provider is anonymous and keeps your prompts. Both of those things are on the same page, and only one of them is getting shared.
What actually happened
OpenRouter lists Ox Alpha under its stealth namespace as "a reasoning model designed for coding, sustained agentic work, and production workloads." The specs are real: 1,048,576 tokens of context, 131,072 max completion tokens, text plus image and video input, function calling, JSON mode. Input and output both price at $0.
The provider line is the part to read: the model is "developed and operated by a third-party provider who has chosen to remain anonymous during this preview." And on data: "prompts and completions for this model are retained by the provider and are not used for training." Everything else falls under OpenRouter's Stealth Model Terms.
TechCrunch reported on August 23 that nobody outside the deal knows who built it. The guessing has run from Chinese labs to an unreleased Microsoft model. None of it is confirmed, which is the whole point of a stealth launch — the lab gets clean benchmark signal and a wave of free evaluation traffic without its name attached to the misses.
Why an anonymous model provider matters for your business
"Not used for training" is not "not stored." Those are two different promises, and the listing only makes the first one. Retention means your prompts sit somewhere, under a jurisdiction you cannot name, operated by a company you cannot name. If your coding agent's context window contains customer records, private repos, or an API key someone pasted into a config file, that is where it went.
Free preview pricing is a clock, not a price. Stealth models exist to collect evaluation data. When the preview closes it closes without notice, and anything you built on top of that endpoint stops working mid-sprint. Fine for a bake-off. Not fine for a job that runs nightly.
Test it the right way. Run Ox Alpha against your hardest refactor with synthetic or public code, score it, write the number down. That is real signal, and it costs you nothing. Then decide whether a named provider with a signed agreement is worth the difference.
- Ox Alpha lists 1,048,576-token context, 131,072 max output, free input and output on OpenRouter
- The provider is anonymous "during this preview" — no name, no jurisdiction, no contract you can point at
- Prompts and completions are retained by that provider; the listing only promises they are not used for training
- Nobody has confirmed who built it; the speculation is speculation
- Benchmark it with data you would publish, not with data you would not
Evaluate every model. Depend on the ones that signed something. We build inference layers where the model is a config value and the data boundary is a policy, so a stealth preview is a test run and never a dependency. See how we build AI systems.
Sources: OpenRouter Ox Alpha model page, TechCrunch.
- #ox-alpha
- #openrouter
- #ai-coding
- #vendor-risk
- #llm
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
TrueForge: an MIT agent harness that runs 30-75% cheaper
TrueFoundry open-sourced TrueForge, an MIT-licensed agent harness. Benchmarked at $2.90 vs $11.80 against Claude Managed Agents. The loop is the part you should own.
Read itQwen-UI-Agent tops GUI benchmarks. No weights yet.
Alibaba's Qwen-UI-Agent posts 82.1% on MobileWorld and 79.5% on OSWorld-Verified. The downloadable GUI agent is still the 2B and 8B predecessor.
Read it