OX Alpha API: Try the 1M-Token Stealth Model Free Today
Felo API now serves OX Alpha, a stealth frontier reasoning model with a 1M-token context. Free per million tokens for a limited time. Try it now.
Two days ago, a frontier model appeared, and nobody would claim it. It had a million-token context window, frontier-grade specs, and a "Stealth" label where the maker's name should be. Industry watchers called it fun; developers called it worth testing.
Today, that same model is available as an API, and it is free per million tokens, at least for now.
You do not need a special invite or a big AI budget. Just open the model page, or make one API call with an API key.
What Is OX Alpha?
OX Alpha is a reasoning model built for long-horizon work: efficient coding, sustained agentic tasks, and production-grade workflows. It surfaced on August 20 under a "Stealth" label, with no maker attached, so nobody can prove who built it. The guesses are loud: a Chinese lab, a rebranded open model, a small team with a very large cluster. None of it is confirmed.
What is confirmed does the talking:
| Feature | Details |
|---|---|
| Context window | 1,048,576 tokens (a whole codebase, or a research stack, in one pass) |
| Max output | 128K tokens in a single response |
| Inputs | Text, images, and video (outputs text) |
| Reasoning | Explicit reasoning episodes, tuned for long-horizon tasks |
| Tool use | Function calling for agentic workflows |
| Output modes | Tool calls, JSON mode, and streaming |
A model that takes a million tokens at once is not a bigger autocomplete. It holds your entire project in its head: the code you wrote last month, the papers you are comparing, the stack trace from yesterday's crash. Then it reasons over all of it in a single pass.
Two caveats, stated plainly. The benchmark hype around OX Alpha is real but unaudited: a viral claim about a DeepSWE coding win came from a 10-question test, and an in-house ranking putting it second on Kingbench is a single developer's numbers, not a verified leaderboard. The maker is anonymous. If you are picking a model for a production stack, read the benchmarks as a rumor and verify with your own workload.
Felo API Platform: One Key, Every Model
OX Alpha is served through the Felo API Platform, a unified gateway that speaks the API dialects your tools already use. Point your OpenAI SDK at one base URL, or your Anthropic client at another, and the platform routes you to whichever model the request names.
- OpenAI-compatible surface:
POST https://openapi.felo.ai/api/v1/chat/completions. Set the OpenAI SDK's base URL tohttps://openapi.felo.ai/api/v1and it works unchanged. - Anthropic-compatible surface:
POST https://openapi.felo.ai/api/v1/messages. Claude-style clients set their base URL tohttps://openapi.felo.ai/api. - A Responses surface at
.../api/v1/responsesfor agent and workflow clients.
Alongside OX Alpha, the platform carries Claude, DeepSeek, GPT, and Grok models. That means one account, one API key, one dashboard, and no per-model vendor dance. For OX Alpha specifically, you get a dedicated model page with the full spec, live pricing, and a drop-in quick start.
How to Try OX Alpha Free in Under a Minute
The fastest path takes about thirty seconds, and you do not need to write a line of code:
- Open https://openapi.felo.ai/models/stealth/ox-alpha, the model's public page on the Felo API Platform.
- Click Open Playground to launch a chat at playground.felo.ai.
- Test the model with your own question.
When you want to wire it into your own code, create an API key from your Felo API Platform account, then set the environment variable and go:
export FELO_API_KEY="YOUR_API_KEY"
A minimal request, cURL edition:
curl -N https://openapi.felo.ai/api/v1/chat/completions \
-H "Authorization: Bearer $FELO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "ox-alpha",
"stream": true,
"messages": [{"role": "user", "content": "Hello"}]
}'
With the OpenAI SDK, the change is three lines:
from openai import OpenAI
client = OpenAI(base_url="https://openapi.felo.ai/api/v1", api_key=FELO_API_KEY)
reply = client.chat.completions.create(model="ox-alpha", messages=[{"role": "user", "content": "Why is my test flaky?"}])
That model ID is the string used everywhere on the platform: ox-alpha.
What It Is Good For
The platform's model card is explicit about what it's "best for": long-horizon coding, complex reasoning, agent workflows, and multimodal tasks. Three practical examples:
The everything-at-once coding question. Paste the crash log, the failing test fixture, and the pull requests that touched the offending file into one request and ask what is actually wrong. The million-token window keeps the whole story visible instead of whatever a summary box still remembers.
The agentic chain. OX Alpha supports function calling and streams tool calls, which makes it a natural fit for multi-step agents. Agents built on Claude Code or Codex can call it through the platform's Anthropic- and OpenAI-compatible surfaces without a custom integration.
The multimodal thread. One context holds an image, a video clip, and a written spec. Ask the model to reconcile them: the screenshot says one thing, the spec another, and the output resolves the contradiction.
Because the model is reasoning-capable, the platform notes you should preserve the returned reasoning metadata when you continue a conversation, and the optional X-Request-Id header can speed up troubleshooting.
The Same Model, Two Ways to Reach It
OX Alpha is now reachable in the browser too. Felo Search already lists it in its model selector (felo.ai/search, or jump straight to it here: felo.ai/search?search_model=ox-alpha) for question-and-answer work. The API you are reading about is the same model for the work that needs code: pipelines, agents, batch jobs, and the inside of your own product.
That split is worth keeping in mind. If the question is yours and the answer is for the screen, the search box is the fastest path. If the task runs on a schedule, signs your code, or ships inside a product, the API is where it belongs.
Read With Caution, Try With Confidence
The right posture toward this model is skepticism about the claims, and curiosity about the cost. During this launch window, the OX Alpha API is listed with free per-million-token pricing on both input and output. That makes first-hand verification genuinely cheap. Generally speaking, the platform's launch pricing is where new models are easiest to try without a budget review.
That is the honest way to judge a mystery model anyway: bring the hardest question you already know the answer to, and see what it comes up with. If it keeps up, the absence of a vendor logo stops mattering.
Try OX Alpha Before the Free Window Closes
A stealth model with a million-token context showed up out of nowhere two days ago. It is now one model ID and one API key away. And for now, it is free. Launch windows have a way of closing.
If you just want to poke at it in the browser, the playground is waiting. If you want it in production, the endpoints and quick start have been quoted in full above.
Try OX Alpha Free on Felo API →
FAQ
What is the OX Alpha model ID?
ox-alpha. It is the same string in the playground, in cURL, in every SDK, and in the models directory at openapi.felo.ai/models/stealth/ox-alpha.
Do I need my own OpenAI or Anthropic account?
No. The platform exposes OpenAI-compatible and Anthropic-compatible surfaces, so existing SDKs and agent tools work against Felo endpoints with a single Felo API key. Bring your clients, not your accounts.
Is OX Alpha really free?
The model page lists free per-million-token pricing on input and output during this launch window. Launch pricing does not survive launch windows, so check the model page before you click.
Should I trust those benchmarks?
Treat them as a rumor from a very well-spoken stranger. The samples are small, the leaderboard is unaudited, and the maker is anonymous. The cheap way to resolve it is your own question, and this is a free cheap way.
Note: benchmark claims referenced here are preliminary and unaudited. Pricing shown reflects the platform's published launch listing and may change; verify current rates on the model page.
This post is also available in 简体中文, 日本語, 한국어, 繁體中文, हिन्दी, Français, العربية, Русский, اردو, Bahasa Indonesia, Deutsch, Tiếng Việt, Türkçe, Italiano, ไทย, Español, বাংলা and Português.