Ox Alpha

Anonymous frontier model for coding and agentic work with 1M token context.

Context window

1,048,576 (1M tokens)

Input / 1M tokens

$0.00

Output / 1M tokens

$0.00

Provider

Anonymous (Stealth)

Data verified 2026-08-23

Ox Alpha is a reasoning model designed for coding, sustained agentic work, and production workloads. It features a 1,048,576-token context window with multimodal input (text, image, video), tool calling, and structured output support. Released anonymously through OpenRouter on August 20, 2026, with free access during preview period.

Capability index

Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.

Coding

Reasoning

Math

Multimodal

Long context

100

Speed

Cost efficiency

100

How to access it

Available on OpenRouter and OpenCode under the provider name 'Stealth.' Create an API key and call stealth/ox-alpha through the standard chat-completions endpoint. Free during preview week (approximately until August 27, 2026).

Get access →Documentation →

Strengths

✓1-million token context window for large codebases and long-horizon tasks
✓Multimodal input (text, image, video)
✓Strong early community performance on coding benchmarks (80% Pass@1 on DeepSWE)
✓Tool calling and structured JSON output support
✓No prompt/completion training without retention claims by OpenCode

Best for developers who...

Long-horizon software engineering tasksComplex multi-file code modificationsAgent workflows with sustained reasoningRepository-scale analysis and refactoring

When to choose it (and when not to)

Reach for Ox Alpha when...

→When you need a very large context window for code-heavy work
→For testing frontier coding capabilities at zero cost
→During the free preview window (approximately until August 27, 2026)

Look elsewhere if...

✕Production workloads with sensitive data (provider is anonymous and retains prompts/completions)
✕Workflows requiring a named vendor or signed data processing agreement
✕Long-term deployments (pricing and availability post-preview are unconfirmed)

How to use it

›Include full codebase context - the 1M window is designed for repository-scale work
›Use tool_choice to direct the model to function calling for structured tasks
›Test real engineering workflows (bug fixes, feature additions) rather than simple chat prompts

Quickstart

Python

import openai

client = openai.OpenAI(
    api_key="your-openrouter-key",
    base_url="https://openrouter.ai/api/v1"
)

response = client.chat.completions.create(
    model="stealth/ox-alpha",
    messages=[{"role": "user", "content": "Fix the bug in this code..."}]
)
print(response.choices[0].message.content)

Use your OpenRouter API key. Free during preview; pricing post-preview is unannounced.

API model id: stealth/ox-alpha

Benchmarks

Benchmark	Score	Notes
DeepSWE (community test subset)	80% Pass@1	User-run test on 10 tasks, unaudited; outperformed GPT-5.6-SOL (~52%) and Claude Opus 5 (~65%) on tested subset but sample size is small and statistical noise is high.

Source: Community testing (Medium, The Geek Insights)

Compare Ox Alpha

Compare Ox Alpha with any other model

Build a comparison →All model comparisons →

Learn the concepts

Inference Context Window

← All AI models