Ox Alpha is a free stealth reasoning model aimed at coding, long-running agent work, and production-style tasks. The confirmed version has a 1,048,576-token context window, accepts text, images, and video, returns text, supports tool calling, and is available through both OpenCode Zen and OpenRouter under different model IDs.
Quick answer: Ox Alpha is worth trying while access is free, especially for repository-scale analysis and multimodal review. Treat it as a preview rather than a production dependency: its developer is still undisclosed, the free window is temporary, and access policies depend on the route you use.
Ox Alpha can inspect code, images, and video, but its output is still text. If your workflow extends from analysis into AI creation, the SeeAPI AI creation platform is a broader starting point for exploring the available models and creative workflows.
Ox Alpha at a Glance
Item | Confirmed information |
|---|---|
Primary use | Coding, sustained agent work, complex reasoning, and production workloads |
Context window | 1,048,576 tokens |
Input | Text, images, and video |
Output | Text |
Maximum output | Up to 131,072 tokens on the listed OpenRouter route |
Tool support | Tools, tool choice, structured responses, and reasoning controls |
Price during preview | Free on the currently listed OpenCode and OpenRouter routes |
Developer | Not publicly disclosed |
The large context window is the headline, but the more useful story is the combination of long context, multimodal input, mandatory reasoning, and tool support. That makes Ox Alpha a candidate for tasks that require more than a short code completion.
What Is Ox Alpha?
Ox Alpha is a stealth model: the model is available for public use while its developer remains anonymous. OpenCode describes it as a limited-time free model and lists it among the models served through OpenCode Zen. OpenRouter separately describes it as a reasoning model for coding, sustained agentic work, and production workloads.
This does not mean OpenCode or OpenRouter created the underlying model. They are access routes. Until the model provider reveals itself, claims that Ox Alpha is a new GLM, MiMo, MiniMax, Qwen, or another named model remain speculation.
The naming also changes by route:
Access route | Model ID | Endpoint style | Route-specific note |
|---|---|---|---|
OpenCode Zen |
| OpenAI-compatible Chat Completions | OpenCode describes this route as zero retention and not used for training |
OpenRouter |
| OpenAI-compatible Chat Completions | Free preview route; check the current provider policy before sending sensitive data |
How to Use Ox Alpha on OpenCode
OpenCode Zen lists Ox Alpha Free with the model ID x-preview-f-free. In OpenCode, connect a Zen API key and select the model from the model list. The same route can be called through the OpenAI-compatible Chat Completions endpoint documented by OpenCode Zen.
curl https://opencode.ai/zen/v1/chat/completions \
-H "Authorization: Bearer $OPENCODE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "x-preview-f-free",
"messages": [
{
"role": "user",
"content": "Review this migration plan and identify the first three risks."
}
],
"max_tokens": 4096
}'Use a meaningful output allowance. Ox Alpha has mandatory reasoning, so an extremely small max_tokens value can leave too little room for the visible answer after reasoning tokens are consumed.
How to Call Ox Alpha Through OpenRouter
OpenRouter exposes the same public name as stealth/ox-alpha. Its current model record lists free input and output, a 1M-token context window, text/image/video input, text output, and tool-related parameters. You can inspect the current route on the official Ox Alpha model page.
curl https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "stealth/ox-alpha",
"messages": [
{
"role": "user",
"content": "Create a safe test plan for this repository refactor."
}
],
"reasoning": { "effort": "high" },
"max_tokens": 4096
}'Parameters and availability can change during a stealth preview. Read the live model record before building a long-lived integration around the current defaults.
What the 1M Context Window Actually Helps With
A one-million-token limit does not automatically make every answer better. It gives a model room to receive a much larger working set, but the quality of that working set still matters.
Useful experiments include:
Repository mapping: provide architecture notes, selected source files, tests, and issue context so the model can trace relationships before proposing a change.
Long-running agent sessions: preserve more tool results, decisions, and intermediate findings without compacting every few turns.
Multimodal debugging: combine code with screenshots or a short product recording when the bug involves both interface behavior and implementation.
Document synthesis: compare specifications, migration notes, logs, and implementation plans inside one controlled task.
Do not upload an entire repository simply because the limit is large. Start with a clear objective and the smallest useful evidence set, then expand when the model identifies a real gap.
Privacy and Zero Data Retention: The Route Matters
A model name is not a privacy policy. Treat retention, training use, hosting region, and account controls as properties of the route you are using.
OpenCode says its Ox Alpha Free provider follows a zero-retention policy and does not use the data for model training. That is a meaningful reason to test through the OpenCode route when private code is involved, but normal security practice still applies: remove secrets, minimize customer data, and confirm the current policy before sensitive work.
Do not automatically carry that statement over to every gateway that offers Ox Alpha. The model ID, provider, privacy terms, and availability may differ even when the public model name is the same.
Who Made Ox Alpha?
The developer has not been publicly confirmed. Community comparisons may produce useful hypotheses, but response style, self-identification prompts, tokenizer behavior, or one matching answer pattern cannot prove model ownership.
Confirmed | Still unknown |
|---|---|
Ox Alpha is a stealth reasoning model | The underlying laboratory or company |
It is optimized for coding and agent work | Whether it is a preview of a named future release |
It accepts text, images, and video | Model size, architecture, and training data |
It is currently free on listed routes | Long-term price and availability |
This distinction is important for searchers and developers. A useful Ox Alpha guide should help people use the model today without turning speculation into a headline.
Five Tests to Run During the Free Window
Repository orientation: ask for a map of modules, dependencies, and likely change points before requesting code.
Long-context consistency: provide a structured design document and test whether later answers still respect early constraints.
Tool-call reliability: give the model one allowlisted read-only tool, validate its arguments, and inspect the raw response.
Screenshot-to-code diagnosis: attach a UI screenshot with the relevant component code and ask for evidence-ranked causes.
Video workflow review: provide a short recording of a failing interaction and ask the model to create a reproducible debugging checklist.
Run each test with a narrow success condition. A free preview can tell you whether the model deserves deeper evaluation, but it cannot guarantee stable production behavior after the provider, price, or model identity changes.
If you want a local rather than hosted agent experiment, the Qwen 3.8 27B local agent guide covers a different path built around controlled local inference.
Limits to Know Before Production
The identity is undisclosed. You cannot evaluate the provider's long-term roadmap or governance from the model name alone.
The free period is temporary. OpenCode announced a one-week free window, and either route can change limits or availability.
Reasoning is mandatory on the OpenRouter record. Budget output tokens accordingly and watch total latency.
Multimodal input is not media generation. Ox Alpha analyzes images and video but returns text.
A 1M limit is not a recommendation to use 1M tokens. Larger prompts increase cost, latency, and the amount of irrelevant context the model must navigate when the preview becomes paid.
Is Ox Alpha Worth Trying?
Yes—if your goal is to evaluate a long-context coding or agent model while access is free. Its confirmed combination of a 1M context window, multimodal input, tool support, and mandatory reasoning makes it more interesting than a generic free chat model.
Keep the experiment reversible. Use a route-specific model ID, avoid sensitive data unless the selected route's policy is appropriate, log prompts and outcomes, and keep another model available as a fallback. The useful result is not discovering the anonymous developer; it is learning whether Ox Alpha handles your real workflow reliably enough to revisit after the preview.
Once Ox Alpha has helped plan, inspect, or troubleshoot the workflow, continue from SeeAPI's creation workspace and choose the model and medium that fit the final asset.
Frequently Asked Questions
Is Ox Alpha free?
It is currently listed as free on OpenCode Zen and OpenRouter. Preview pricing and limits can change, so verify the live model record before a large run.
How long will Ox Alpha remain free?
OpenCode announced on August 21, 2026 that its free access window would run for the following week. OpenRouter availability may follow a different schedule.
Does Ox Alpha support images and video?
Yes. The current model record accepts text, image, and video input, but its output is text.
Who created Ox Alpha?
The developer has not been publicly disclosed. Current claims linking it to a specific laboratory are unconfirmed.
Can I use Ox Alpha through an API?
Yes. OpenCode Zen lists an OpenAI-compatible Chat Completions endpoint with the model ID x-preview-f-free, while OpenRouter uses stealth/ox-alpha.
