OXAlpha
Open chat
Stealth model · origin unknown · free preview

Meet Ox Alpha.

The reasoning model that appeared out of nowhere — built for code, long-horizon agents, and a million tokens of context. This is an independent field guide: what it is, what we tested, and how to use it.

Start chatting →Get the API
1M
tokens of context
1,048,576 exactly
131K
max output tokens
long generations
3
input modalities
text · image · video
$0
cost in stealth
free while previewed
01

Reasoning, always on

Ox Alpha thinks before it answers. Reasoning is mandatory and streams as a visible chain of thought, with three effort levels — low, high, and max. You see the working, not just the conclusion.

Visible thinking
Effort control (low/high/max)
Reasoning continues across turns
02

A million tokens of context

A 1,048,576-token window means you can drop an entire codebase, a book, or a year of logs into a single prompt — no chunking, no retrieval plumbing, no lost-in-the-middle gymnastics.

Whole-repo reasoning
Long-document analysis
128K tokens of output headroom
03

Built for real work

Long-horizon coding, tool / function calling, and multimodal input. It's designed for sustained agentic workflows, not one-shot party tricks.

Tool calling
Vision (image + video in)
Agentic, long-horizon tasks

What people build with it

All use cases →
AI for coding AI for customer support AI for content writing AI for data analysis AI for research AI for summarizing documents AI for email AI for studying and learning

Ox Alpha at a glance

The specs that matter, verified against the live API.

Context window1,048,576 tokens (1M)
Max output131,072 tokens
InputText · Image · Video
ReasoningMandatory · low / high / max
Tool callingSupported
Price$0 during stealth preview
Model idstealth/ox-alpha
AccessOpenRouter API

What we actually tested

Leaderboard claims are easy to make and hard to trust for a model with no named creator. So instead, here are first-hand checks against the live API.

Counts letters correctly ('strawberry' → 3 r's)✓ pass
Function/tool calling returns well-formed tool_calls✓ pass
Streaming tokens + reasoning over SSE✓ pass
Prompt caching active (cached_tokens > 0)✓ pass
Strict response_format json_schema enforcedpartial — returns prose; use tool calls
Cost during stealth$0 across all test calls

FAQ

Who made Ox Alpha?

Nobody knows publicly. It appeared on OpenRouter under the 'Stealth' provider with its origin masked. That anonymity is part of why it's called a stealth model.

Is it really free?

During the stealth period, yes — every call in our testing returned a cost of $0. That can change without notice once the preview ends, so don't build a business on the price staying zero.

What is it good at?

Coding, long-horizon agent work, and anything that benefits from a huge context window plus real reasoning. It also accepts images and video as input.

Should I send sensitive data?

No. It's a preview model routed through a shared stealth pool that may log traffic. Keep secrets, credentials, and personal data out of your prompts.

How do I use it?

Through the OpenRouter chat-completions API with model 'stealth/ox-alpha'. See the API guide on this site for a copy-paste quickstart.

Try it in your browser

A live chat with streaming answers and visible reasoning — free while Ox Alpha is in stealth.

Open chat →Read the API guide