Ox Alpha Wiki
Ox Alpha is a free stealth reasoning model built for coding, sustained agentic work, long-horizon software engineering and multimodal AI workflows.
Ox Alpha Resources
Everything you need to master the Ox Alpha stealth model: specs, API, pricing, coding workflows, and identity tracking
Latest Updates
Discover the newest guides, tips, and content
Ox Alpha 1 million token context: Setup Guide & Limits
Learn what Ox Alpha's 1 million token context supports, where to access it, and how to evaluate its coding and agentic performance safely.
Ox Alpha 1m context: Setup Guide, Limits & Best Uses
Learn how Ox Alpha's 1M context, multimodal input, output limits, privacy claims, and preview access shape practical AI workflows.
Ox Alpha agentic coding: Setup Guide & Best Practices
Learn how Ox Alpha supports agentic coding, long-horizon software work, multimodal prompts, API setup, testing, and production safeguards.
Ox Alpha ai model: API Setup Guide & Benchmarks
Learn what Ox Alpha is, how to access it through OpenRouter, and how to evaluate its context, speed, pricing, and agentic coding fit.
Ox Alpha api: Setup Guide, Limits & Best Practices
Learn how to use the Ox Alpha api through OpenRouter, including setup, multimodal inputs, context limits, pricing, and practical workflow tips.
Ox Alpha artificial analysis: Capabilities & Benchmarks
Ox Alpha artificial analysis covering context, multimodal input, coding benchmarks, 3D generation, and practical evaluation advice.
Ox Alpha benchmark results: DeepSWE Scores & Comparisons
Review Ox Alpha benchmark results, including its reported 80% DeepSWE score, model comparisons, task performance, and evaluation limits.
Ox Alpha benchmark: Early Scores, Context, and Model Comparison
Review the reported Ox Alpha benchmark results, context window, multimodal features, reliability limits, and comparison with competing AI models.
Ox Alpha chat completions: Step-by-Step API Setup Guide
Learn how to configure Ox Alpha chat completions, authenticate requests, set reasoning options, control generation, and read responses.
Ox Alpha complex reasoning: Setup Guide & API Tips
Learn how Ox Alpha handles complex reasoning, coding, visual context, API setup, performance, privacy, and practical testing workflows.
Ox Alpha computer use: Setup Guide, Tests & Limits
Learn how Ox Alpha computer use works, what its multimodal agent can do, and how to test it responsibly with clear benchmarks and limits.
Ox Alpha evals: Benchmarks, Setup & Safety Tips
Explore Ox Alpha evals, reported benchmarks, practical testing methods, access options, and privacy precautions for the anonymous AI model.
Ox Alpha Model Specifications
Ox Alpha launched on August 20, 2026 under OpenRouter's stealth namespace. Its standout feature is a 1,048,576-token context window, allowing developers to process very large codebases, documents, logs, and long-running agent histories in a single context.
Ox Alpha
stealth/ox-alpha
OpenRouter Stealth
August 20, 2026
1,048,576 tokens
131,072 tokens
Text, Image, Video
Text
Reasoning-capable
Coding, agentic workflows, repository analysis, long-context tasks, and production workloads
Available through the OpenRouter API
Why Ox Alpha Stands Out
- A 1M-class context window large enough for whole repositories, long documents, and full agent histories.
- Multimodal text, image, and video input with up to 131,072 output tokens per response.
- Reasoning-capable and positioned for coding, agentic workflows, and production workloads.
Ox Alpha API Setup Guide
Ox Alpha uses the model identifier stealth/ox-alpha. Developers can call it through OpenRouter's OpenAI-compatible chat completions endpoint, making it straightforward to integrate into applications that already use OpenAI-compatible APIs.
Create an OpenRouter API Key
Create an OpenRouter API key and store it securely as an environment variable such as OPENROUTER_API_KEY.
export OPENROUTER_API_KEY=sk-or-...Use the Ox Alpha Model ID
Set the model parameter to stealth/ox-alpha in every request.
"model": "stealth/ox-alpha"Send a Request with curl
Call the OpenRouter chat completions endpoint and provide your API key in the Authorization header.
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":"Explain this codebase architecture."}]}'Use Ox Alpha with Python
The OpenAI Python SDK connects to OpenRouter by changing the base URL while keeping the standard chat completions interface.
from openai import OpenAI
client = OpenAI(
base_url="https://openrouter.ai/api/v1",
api_key=os.environ["OPENROUTER_API_KEY"]
)
response = client.chat.completions.create(
model="stealth/ox-alpha",
messages=[{"role": "user", "content": "Review this implementation."}]
)
print(response.choices[0].message.content)Use Ox Alpha with TypeScript
The OpenAI JavaScript SDK uses the same OpenRouter-compatible endpoint with the stealth/ox-alpha model identifier.
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://openrouter.ai/api/v1",
apiKey: process.env.OPENROUTER_API_KEY
});
const response = await client.chat.completions.create({
model: "stealth/ox-alpha",
messages: [{ role: "user", content: "Debug this function." }]
});
console.log(response.choices[0].message.content);Enable Streaming
Set stream to true when you want tokens returned incrementally instead of waiting for the complete response.
"stream": trueOx Alpha Pricing and Free Access
The stealth/ox-alpha listing is positioned as a free-access model on OpenRouter. Its pricing and availability can change as the stealth preview evolves, so the OpenRouter model page remains the main source for the current access status.
Free
$0
$0
stealth/ox-alpha
1,048,576 tokens
OpenRouter
OpenRouter API key
Free preview access through OpenRouter
Subject to OpenRouter account, provider, and free-model rate limits
Free Access Notes
- Free preview pricing applies while the stealth rollout remains active on OpenRouter.
- Free-model requests are subject to per-model and per-account rate limits.
- If the stealth preview ends, the OpenRouter model page will list the updated pricing route.
- Anonymous stealth providers on OpenRouter retain prompts and completions for abuse monitoring but do not use them for training.
Ox Alpha Coding and Agentic Workflows
Ox Alpha is built for workflows where the model must keep substantial project context available while reasoning across many files or multiple task stages. OpenRouter's API and tool-calling support make it possible to place the model inside autonomous or human-supervised coding agents.
Repository Analysis
Provide large portions of a repository, architecture notes, configuration files, and documentation so Ox Alpha can reason about relationships across the project.
Best for: Codebase onboarding, architecture reviews, dependency analysis, and refactoring plans
Long-Horizon Coding
Keep requirements, implementation details, previous changes, and test results in context while working through multi-stage engineering tasks.
Best for: Feature development, migrations, large refactors, and multi-file changes
Debugging Workflows
Combine source code, logs, stack traces, configuration, and reproduction steps in one prompt so the model can trace failures across multiple components.
Best for: Bug diagnosis, production incident investigation, and regression analysis
Tool-Calling Agents
Connect Ox Alpha to tools that can search files, inspect repositories, run tests, query services, or perform application-specific actions.
Best for: Coding agents, developer assistants, and automated engineering workflows
Plan-Execute-Review Loops
Have the model first create a task plan, execute individual steps through tools, inspect the results, and revise its next action based on what happened.
Best for: Complex tasks that require multiple dependent actions
Large Documentation Tasks
Use the extended context window to analyze specifications, API references, technical documentation, and source code together.
Best for: Documentation generation, implementation planning, and specification-to-code workflows
Code Review
Supply related files and surrounding implementation context instead of reviewing an isolated diff, allowing broader reasoning about behavior and maintainability.
Best for: Pull request review, security checks, and maintainability analysis
Production Agent Pipelines
Combine structured prompts, tool calls, streaming responses, validation, and application-side controls to integrate Ox Alpha into production workflows.
Best for: Internal developer tools, automation systems, and agentic applications
Ox Alpha 1M Context and Multimodal Guide
Ox Alpha supports a 1,048,576-token context window plus text, image, and video inputs. Its large context makes it suitable for codebases, long documents, multi-file analysis, and extended agent workflows, with text-based responses returned through the OpenRouter API.
1,048,576-Token Context Window
1,048,576 tokens
The large context window can hold extensive prompts, source files, documentation, conversation history, retrieved records, and other supporting material in a single model context.
Large codebases
Long technical documents
Repository-wide analysis
Extended conversations
Large Codebase Analysis
Multi-file context
Developers can provide many related source files, configuration files, logs, tests, and documentation together so Ox Alpha can reason across dependencies instead of analyzing files independently.
Repository exploration
Cross-file debugging
Refactoring
Architecture analysis
Long Document Processing
Long-context analysis
The context window can be used for lengthy reports, specifications, transcripts, research material, contracts, technical manuals, and collections of related documents.
Document summarization
Information extraction
Cross-document comparison
Question answering
Text Input
Supported
Standard text prompts, source code, structured text, conversation messages, and long-form documents can be supplied as model input.
Chat
Coding
Reasoning
Extraction
Image Input
Supported
Image content can be included alongside text prompts for multimodal analysis workflows.
Screenshot analysis
Visual document understanding
UI analysis
Image-based reasoning
Video Input
Supported
Ox Alpha exposes video as an accepted input modality, allowing applications to combine video material with textual instructions.
Video understanding
Scene analysis
Content review
Multimodal agents
Text Output
Text
Model generations are returned as text, including natural-language answers, code, tool-related responses, and structured text when the appropriate API features are used.
Application responses
Generated code
Reports
Structured data
Extended Agent Context
Long-running context
Large context capacity is useful for agent workflows that accumulate instructions, tool results, retrieved documents, code, and conversation history across complex tasks.
Coding agents
Research agents
Automation
Production workflows
Ox Alpha Reasoning Settings
Ox Alpha is a reasoning-oriented model and can use OpenRouter reasoning controls for supported requests. Adjust reasoning effort according to task complexity, latency requirements, and token usage: higher effort for complex coding and planning, lower settings for rapid tests and straightforward tasks.
Allocates greater reasoning effort to difficult multi-step problems, planning, analysis, and tasks where solution quality matters more than response speed.
{
"reasoning": { "effort": "high" }
}Useful for repository-level debugging, architecture decisions, difficult implementation tasks, refactoring, and problems requiring several dependent reasoning steps.
{
"reasoning": { "effort": "high" }
}A balanced configuration for everyday programming, technical questions, data transformation, application logic, and moderate reasoning workloads.
{
"reasoning": { "effort": "medium" }
}Reduces reasoning effort for simple prompts, API integration checks, formatting tasks, lightweight transformations, and workflows where lower latency is preferred.
{
"reasoning": { "effort": "low" }
}Provides a token-based reasoning budget when a developer wants more direct control over how much reasoning capacity is allocated to a request.
{
"reasoning": { "max_tokens": 4000 }
}Excludes returned reasoning information while still allowing the model to perform reasoning internally, which is useful when an application only needs the final response.
{
"reasoning": { "effort": "high", "exclude": true }
}OpenRouter can expose reasoning-related response data through reasoning detail fields for workflows that preserve or process supported reasoning metadata.
{ "response_field": "reasoning_details" }Ox Alpha Tool Calling and Structured Outputs
Tool calling lets an application describe external functions that the model can select during a task, while structured outputs constrain generated data into formats that software can consume reliably. Together they allow Ox Alpha to operate as the reasoning layer inside larger application workflows.
Define Available Tools
toolsSend function definitions describing operations that the model is allowed to request, including the function name, purpose, and expected arguments.
Control Tool Selection
tool_choiceConfigure how tool selection is handled so an agent can allow automatic function selection or guide the model toward an appropriate tool for a workflow.
Execute the Requested Function
tool_callsWhen Ox Alpha requests a function, the application reads the generated tool call, validates its arguments, executes the matching external function, and captures the result.
Add the external function result back to the conversation as tool output so the model can use the returned data to continue reasoning and produce the next response.
Request JSON Responses
response_formatUse response-format controls when application code needs machine-readable JSON rather than an unrestricted natural-language response.
Apply a JSON Schema
response_format.json_schemaStructured outputs can constrain generated data to a defined JSON schema, giving backend applications predictable field names, data types, and object structure.
Build Agent Loops
tools + structured outputCombine reasoning, tool calls, returned tool results, and schema-constrained final output to build production agents that can perform actions and return data in an application-ready format.
Ox Alpha Benchmarks and Model Identity
The most useful way to compare Ox Alpha is to separate its exposed OpenRouter specifications from discussion about the model behind the stealth identifier. The API-facing model name, context size, modalities, reasoning features, and supported production capabilities can be tracked independently from community attempts to identify its underlying family.