- Ox Alpha price: OpenRouter currently lists both input and output usage at $0 during the preview.
- Context window: The model supports a 1M-token context, suited to long coding sessions and large projects.
- Primary use: Ox Alpha targets coding, sustained agentic work, reasoning, and visual-context workflows.
- Provider status: The third-party provider remains anonymous while the model is in stealth preview.
- Best practice: Review retention terms and monitor availability before using it for sensitive production workloads.
Ox Alpha Price and Current Availability
As of August 22, 2026, the Ox Alpha price shown by OpenRouter is free for both prompt and completion tokens. The listing reports a weighted average input price of $0 per million tokens and a weighted average output price of $0 per million tokens. This makes the model useful for testing, prototyping, coding experiments, and agent evaluations without a displayed token charge during the preview.
The listing identifies Ox Alpha as stealth/ox-alpha and states that OpenRouter routes requests directly to one provider. The provider is not publicly named, so the free rate should be treated as a preview condition rather than a permanent commercial promise. Pricing, availability, or access rules may change after the anonymous provider ends its preview period.
View the current Ox Alpha API pricing and provider page on OpenRouter before starting a significant workload.
Video Highlights:
- Ox Alpha is tested across front-end design, coding agents, 3D scenes, and visual tasks.
- The model is evaluated against several other reasoning and coding systems.
- The investigation focuses on tool calls, revisions, task behavior, and possible model lineage.
- The testing suggests strong front-end performance with more mixed results in some 3D workflows.
| Pricing item | Current listing | What it means |
|---|---|---|
| Input tokens | $0 per million | No displayed charge for submitted prompt tokens during the preview |
| Output tokens | $0 per million | No displayed charge for generated completion tokens during the preview |
| Provider count | 1 | Requests are forwarded directly to the listed stealth provider |
| Model access | OpenRouter | Use the model slug stealth/ox-alpha through compatible API tooling |
| Price stability | Preview-dependent | The current free rate may change when the preview or access policy changes |
Use the free preview to benchmark representative tasks, but avoid assuming that today’s $0 rate will remain unchanged for long-term production planning.
What the Free Preview Includes
The OpenRouter listing describes Ox Alpha as a reasoning model designed for coding, long-horizon software engineering, sustained agentic work, and production-oriented workflows. It also supports tasks that combine text with visual context. This positioning makes the model broader than a simple chat assistant: its value depends on how well it can plan, use tools, inspect results, and revise work.
Ox Alpha is listed with a 1M-token context window, which can accommodate large specifications, extended conversations, source repositories, logs, and multi-stage agent sessions. A large context does not automatically guarantee better reasoning, so users should still provide focused instructions and validate generated changes.
The model’s listed modalities include text, images, and video as input, with text output. That combination can support visual debugging, interface analysis, design references, and multimodal coding tasks. Availability for a particular client depends on whether the client correctly passes the supported content format to OpenRouter.
Coding
- Long-form software tasks
- Code generation and revision
- Repository-scale context
Agent Work
- Tool calling
- Iterative inspection
- Multi-step execution
Visual Context
- Image input
- Video input
- UI and design references
Reasoning
- Complex problem solving
- Planning across stages
- Production-oriented workflows
| Capability | Listed or observed detail | Practical use |
|---|---|---|
| Context | 1M tokens | Large codebases, specifications, logs, and extended sessions |
| Input types | Text, images, and video | Visual debugging and multimodal analysis |
| Output type | Text | Explanations, code, plans, and structured responses |
| Intended workloads | Coding and sustained agentic work | Software engineering and tool-assisted workflows |
| Current access | Free preview through OpenRouter | Evaluation before possible pricing changes |
A 1M-token context window is a capacity limit, not a recommendation to paste every available file. Start with the smallest relevant context and add material when the task requires it.
How to Use Ox Alpha Through the API
Ox Alpha uses the OpenRouter model identifier stealth/ox-alpha. OpenRouter presents an OpenAI-compatible interface, so many existing SDK integrations can be adapted by changing the base URL, API key, and model name. Streaming is available and can make long responses easier to monitor.
The basic workflow is straightforward: create an OpenRouter API key, store it in an environment variable, select the Ox Alpha model slug, and send a request. For coding agents, add tools only when needed and define clear boundaries for file changes, shell commands, browser access, or other external actions.
Create an OpenRouter API Key
Sign in to OpenRouter and create an API key from the dashboard. Store it as an environment variable such as OPENROUTER_API_KEY rather than placing the secret directly in source files.
Select the Correct Model Slug
Set the model to stealth/ox-alpha. Using the exact slug prevents accidental requests to a different model or an invalid endpoint.
Send a Focused First Request
Begin with a small coding, reasoning, or visual-context task. Ask for the intended plan, expected files, assumptions, and validation method before allowing broad changes.
Enable Streaming When Useful
Add "stream": true when you want incremental output. Streaming is helpful for tracking long explanations, plans, and agent responses.
Validate the Result
Run tests, inspect generated files, review tool calls, and compare the output against the original requirements. Free access does not remove the need for quality control.
| Parameter | Default or type | Recommended use |
|---|---|---|
max_tokens | Integer | Set an output ceiling appropriate to the task |
temperature | 1 | Adjust when you need more or less response variety |
top_p | 0.95 | Keep the default initially, then tune for specialized generation |
tools | Array | Add only the tools required for the workflow |
tool_choice | String or object | Control whether a tool may or must be called |
response_format | Map | Request structured output when your client supports it |
A minimal request can use the OpenRouter SDK or another compatible client. The important values are the OpenRouter endpoint, your API key, and the stealth/ox-alpha model identifier. For multimodal requests, pass image or video content in the format supported by the API and client library.
Test the connection with a short request first. Once the endpoint works, add streaming, tools, visual inputs, and larger context in separate stages so failures are easier to diagnose.
Performance, Limits, and Privacy Considerations
OpenRouter reports a recent P50 throughput of approximately 23 tokens per second for the Ox Alpha provider and a P50 latency of about 5.30 seconds on the provider listing. The same page reports 99.99% uptime and 99.51% availability over the displayed three-day period. These measurements can vary by location, traffic, prompt size, queue conditions, and tool usage.
The page also reports a tool-call error rate of approximately 2.27% and a cache hit rate near 81.72% for the displayed provider data. Treat these figures as monitoring snapshots rather than guarantees. A coding agent may experience higher end-to-end latency because it performs several calls, waits for tools, reads files, and repeats corrections.
| Metric | Reported value | How to interpret it |
|---|---|---|
| P50 throughput | 23 tok/s | Typical provider-side generation speed shown in the listing |
| P50 latency | 5.30 s | Reported round-trip latency at the listed provider |
| Three-day uptime | 99.99% | Time with at least one provider responding |
| Three-day availability | 99.51% | Portion of requests successfully served |
| Tool-call error rate | 2.27% | Snapshot of tool-related failures in the displayed data |
| Cache hit rate | 81.72% | Reported cache behavior for the provider snapshot |
Privacy requires separate attention. OpenRouter states that prompts and completions for this stealth model are retained by the provider and are not used for training. It also states that other use is governed by the Stealth Model Terms. Retention still means that sensitive information should be handled carefully.
Avoid sending:
- Private credentials, access tokens, or production secrets.
- Personal information that is unnecessary for the task.
- Confidential source code without organizational approval.
- Unreleased business information or regulated data.
- Files that contain hidden keys in comments, configuration, or logs.
The free Ox Alpha preview is not a reason to ignore data governance. Remove secrets, minimize confidential context, and review the applicable Stealth Model Terms before using private workloads.
Best Use Cases and Evaluation Checklist
Ox Alpha appears best suited to workflows where a large context window, tool use, and iterative reasoning matter more than a one-shot answer. A practical evaluation should use the same prompts across several models and score both the final artifact and the process used to produce it.
The available testing material points to strong front-end design results, including animated landing pages and interface work. It also describes successful tool navigation in a Blender workflow, although the final 3D result was less polished than a comparison model. Game-development-style tasks were functional in some cases but visually uneven. These observations support a task-specific evaluation rather than a single overall ranking.
| Task category | Suggested test | Evaluation focus |
|---|---|---|
| Front-end design | Build a responsive animated landing page | Visual polish, layout accuracy, interaction quality |
| Repository coding | Add a feature across multiple files | Planning, safe edits, tests, regression control |
| Multimodal analysis | Inspect a screenshot or short video | Visual accuracy, actionable recommendations |
| 3D workflow | Create or edit a simple scene | Tool reliability, geometry, materials, final presentation |
| Agent loop | Plan, execute, inspect, and patch a task | Recovery behavior, tool calls, and iteration quality |
Before Using Ox Alpha for a Larger Project:
- Confirm the current price and provider status on OpenRouter
- Remove credentials and unnecessary confidential data
- Test a representative task with the exact intended tools
- Check generated code with tests, linting, and manual review
- Record latency, failures, and output quality for comparison
For a fair benchmark, keep the prompt, input files, tool definitions, and success criteria consistent. Measure:
- Final correctness and completeness.
- Number of tool calls and failed calls.
- Time to a usable result.
- Quality of recovery after an error.
- Amount of manual cleanup required.
- Reliability across repeated runs.
Judge Ox Alpha by the workflow you actually need. A model that excels at interface generation may not be the best choice for 3D production, debugging, or highly sensitive code.
Ox Alpha Price FAQ
Q: What is the current Ox Alpha price?
As of August 22, 2026, OpenRouter lists Ox Alpha at $0 per million input tokens and $0 per million output tokens. This rate applies to the displayed preview and may change later.
Q: Is Ox Alpha permanently free?
No permanent free price is confirmed. The model is presented as a stealth preview operated by an anonymous third-party provider, so access and pricing should be checked before each major deployment.
Q: What is Ox Alpha’s context window?
Ox Alpha is listed with a 1M-token context window. It accepts text, images, and video as input and returns text, making it suitable for large coding and visual-context workflows.
Q: Is Ox Alpha safe for confidential source code?
Use caution. OpenRouter states that prompts and completions are retained by the provider and are not used for training, while other use is governed by the Stealth Model Terms. Remove secrets and follow your organization’s data policy.
For the latest price, model slug, provider metrics, and access conditions, check the Ox Alpha OpenRouter listing before sending production requests.