Ox Alpha
Ox Alpha is a stealth multimodal reasoning model with a 1M-token context window, free during preview, designed for coding, agentic workflows, and multimodal tasks.
Overview
What is Ox Alpha
Ox Alpha is a frontier multimodal reasoning model released anonymously on August 20, 2026, via OpenRouter. It offers a 1,048,576-token context window, supports vision and video input, and is optimized for code and sustained agentic work — all at no cost during its preview period.
Running it privately on Venice
Running Ox Alpha on Venice ensures your prompts are anonymized and not stored, preserving your privacy during this experimental phase. With Venice's uncensored access, you can leverage its full capabilities — reasoning, web search, and multimodal input — without surveillance or data retention, ideal for developers testing frontier AI in production workflows.
Assessment
Strengths and limitations
- Massive 1M-token context window enables processing of entire codebases or long documents in a single prompt.
- Multimodal support for text, image, and video input makes it suitable for complex reasoning workflows.
- Free during preview period with no cost for input or output tokens.
- Optimized for code generation and long-horizon software engineering tasks.
- Supports tool use, web search, and structured JSON output for agentic applications.
- Developer and model architecture are anonymous and undisclosed, raising questions about long-term support and transparency.
- No information on parameter count, training data, or model lineage.
- Currently in preview with limited availability and no guaranteed uptime or SLA.
- Not open-source or open-weights, so self-hosting or fine-tuning is not possible.
Capabilities
What it supports
- Tool use / function calling
- Vision (image input)
- Reasoning
- Web search
- Code-optimized
- Structured output (JSON schema)
- Audio input
- Video input
- Multiple image inputs
- Log probabilities
Specifications
Datasheet
- Maker
- Stealth
- Released
- August 20, 2026
- Modality
- Text, image, video input
- Architecture
- Undisclosed
- Parameters
- Undisclosed
- Open weights
- No
- Context window
- 1,048.576K tokens
- Max output
- 131.072K tokens
- Capabilities
- Vision, Function calling, Reasoning, Web search, Code-optimized
- Privacy on Venice
- Anonymized — prompts not stored
- Available on Venice since
- Aug 2026
- License
- Proprietary
API
Call it from your code
Venice exposes an OpenAI-compatible API. Point your base URL at Venice and pass the model id.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "stealth-ox-alpha",
"messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
}'Pricing
What it costs on Venice
Billed per token on Venice: $0 per 1M input tokens and $0 per 1M output tokens.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
Alternatives
How it compares
| Model | Max resolution / input | Strongest at | Open weights | Price (Venice) |
|---|---|---|---|---|
| Ox Alpha | Video input supported | Long-context reasoning, code, agentic work | No | $0 in · $0 out / 1M |
| Claude Fable 5 | Image input | General reasoning, safety | No | $12 in · $60 out / 1M |
| DeepSeek V4 Flash 0731 | Text only | Speed, cost-efficiency | Yes | $0.17 in · $0.35 out / 1M |
| Gemini 3.7 Flash | Image input | Low-latency tasks | No | $1.88 in · $9.38 out / 1M |
Anonymous frontier model with 1M context, free during preview. Ideal for developers testing long-horizon AI agents.
Use cases
What it is good for
- 01Analyzing and generating code across large codebases using full-context awareness.
- 02Multimodal reasoning tasks combining text, images, and video inputs.
- 03Autonomous agent workflows requiring long memory and web search integration.
- 04Prototyping AI applications with zero cost during the preview period.
- 05High-throughput code generation and refactoring with function calling support.
Prompting
Getting better results
Use precise, structured prompts to take full advantage of the 1M-token context window.
Include image or video context when relevant — the model supports multimodal input.
Leverage web search and tool use for real-time data retrieval and external actions.
Specify JSON schema when you need structured output for downstream processing.
Version history
Initial release — 1M context, multimodal, free preview
FAQ
Frequently asked questions
Ox Alpha is a stealth multimodal reasoning model released on August 20, 2026, via OpenRouter. It features a 1M-token context window, supports image and video input, and is optimized for code and agentic workflows. The model is currently free during its preview phase.
Yes, Ox Alpha is free to use during its preview period on Venice and OpenRouter, with $0 pricing for both input and output tokens. This makes it ideal for developers and researchers to experiment at scale without cost.
No, Ox Alpha is not open source or open-weights. It is a proprietary model developed by an anonymous entity under the name 'Stealth,' and there is no public access to its architecture or training data.
The developer of Ox Alpha has not been officially confirmed. It is listed under 'Stealth' on OpenRouter, which states it is only routing the API and not the model provider. Speculation points to Zhipu AI (Z.ai), but no company has claimed ownership.
Yes, Ox Alpha supports both image and video input, enabling multimodal reasoning workflows. This allows users to include visual context in prompts, making it suitable for tasks like UI analysis, video summarization, and image-based code generation.
Ox Alpha has a context window of 1,048,576 tokens — approximately 1 million tokens — allowing it to process extremely long inputs such as entire codebases, lengthy documents, or extended conversations in a single prompt.
Yes, Ox Alpha supports tool use and function calling, enabling integration with external APIs, code execution, and web search. This makes it well-suited for autonomous agent applications and dynamic workflows.
Ox Alpha excels in long-context reasoning and multimodal input at no cost during preview, while Claude Fable 5 offers proven reliability, enterprise safety, and strong general reasoning. Choose Ox Alpha for experimental, high-throughput tasks; Fable 5 for production systems needing trust and consistency.
Run Ox Alpha privately
No prompt logging. No data used for training.