# Ox Alpha

> Public documentation and evaluation notes for Ox Alpha at https://oxalpha.io/.

Ox Alpha is presented on this website as a model for coding, long-context workflows, and agentic work. Model availability, provider routing, parameters, limits, pricing, and commercial terms can change. Verify live provider information before relying on it in production.

## Primary pages

- [Home](https://oxalpha.io/): Product overview, use cases, and public website navigation.
- [Ox Alpha API](https://oxalpha.io/ox-alpha-api.html): OpenAI-compatible Chat Completions integration reference, including cURL, Python, JavaScript, request parameters, response handling, and production guidance.
- [How to Use Ox Alpha](https://oxalpha.io/how-to-use-ox-alpha.html): API-only getting-started guide for the Tokenra route, with Python and cURL examples using `stealth/ox-alpha` and `glm-5.3-flash`.
- [Benchmarks](https://oxalpha.io/benchmarks.html): LiveBench third-party record and evaluation methodology.
- [Model comparisons](https://oxalpha.io/vs/ox-alpha-vs-deepseek-v4-pro.html): Evidence-scoped comparisons with DeepSeek V4 Pro, GLM 5.3, and Claude Fable 5.
- [Ox Alpha Model](https://oxalpha.io/blog/ox-alpha-model.html): A developer guide to the model's coding, long-context, agentic-workflow, API-access, specification, pricing, and evaluation boundaries.
- [Blog](https://oxalpha.io/blog.html): Developer guides for API integration, the Ox Alpha model, LLM evaluation, long-context coding, agent safety, benchmark interpretation, and multimodal input.

## API and provider context

The API documentation uses the model identifier `stealth/ox-alpha` and documents a TokenRa Chat Completions route. The guide also records `glm-5.3-flash` as the ZAI-compatible identifier for the same Ox Alpha model. Use `stealth/ox-alpha` on the documented TokenRa route and `glm-5.3-flash` only where the active ZAI-compatible provider exposes that identifier. Treat TokenRa's live documentation, account flow, availability, parameter support, pricing, and commercial terms as the source of truth for that route.

The [OpenRouter route guide](https://oxalpha.io/blog/ox-alpha-openrouter-guide.html) documents a separate public route for the same model label. A shared label does not establish that OpenRouter and TokenRa use the same provider, model snapshot, request behavior, data handling, availability, or price. Verify each provider route independently.

The public homepage may contain older references to other provider routes. Do not infer that a model label, listing, or route is equivalent to the TokenRa route without direct provider evidence.

## Benchmark context

The benchmark page records a third-party LiveBench public listing for the label `ox-alpha-max` in the `LiveBench-2026-06-25` snapshot. It reports values exactly as displayed by LiveBench, including Overall 69.2 and a displayed cost per successful task of $0.000.

Important limitations:

- The LiveBench label `ox-alpha-max` is not established on this site as equivalent to TokenRa's `stealth/ox-alpha` route.
- The record does not establish a provider route, model snapshot, full request configuration, raw execution artifacts, reproducibility, or a production price.
- Do not use the listing to claim that the TokenRa route achieved these values, that Ox Alpha is free, or that it outperforms another model.

## Model comparisons

The site publishes three focused comparison pages. The DeepSeek and Fable pages use LiveBench-2026-06-25 records; the GLM 5.3 page uses the shared KingBench 3 result and keeps GLM-5.2 LiveBench data as a separate previous-generation reference:

- [Ox Alpha vs DeepSeek V4 Pro](https://oxalpha.io/vs/ox-alpha-vs-deepseek-v4-pro.html)
- [GLM 5.3 vs Ox Alpha](https://oxalpha.io/vs/glm-5-3-vs-ox-alpha.html)
- [Ox Alpha vs Claude Fable 5](https://oxalpha.io/vs/ox-alpha-vs-fable-5.html)

The GLM page reports the supplied KingBench 3 result as GLM-5.3 `91.25% (73/80)` and Ox Alpha `87.5%`. The LiveBench `73.2` value belongs to `GLM-5.2 open` in the `LiveBench-2026-06-25` release and is not a GLM-5.3 result. `ox-alpha-max`, `stealth/ox-alpha`, and `ox-alpha` are treated here as public identifiers for the same Ox Alpha product; benchmark labels and the TokenRa route remain separately disclosed. LiveBench's `Cost per successful task` field is a benchmark metric, not a TokenRa API price, token price, subscription fee, or free-use guarantee.


- [What Is Ox Alpha?](https://oxalpha.io/blog/what-is-ox-alpha.html)
- [Ox Alpha LLM: How to Evaluate a Model Route](https://oxalpha.io/blog/ox-alpha-llm-guide.html)
- [Ox Alpha API Integration Through TokenRa](https://oxalpha.io/blog/ox-alpha-api-guide.html)
- [How to Use Ox Alpha on OpenRouter](https://oxalpha.io/blog/ox-alpha-openrouter-guide.html)
- [Long-Context Coding Workflow](https://oxalpha.io/blog/long-context-coding-workflow.html)
- [Agentic Workflow Safety](https://oxalpha.io/blog/agentic-workflow-safety.html)
- [How to Read LLM Benchmarks](https://oxalpha.io/blog/how-to-read-llm-benchmarks.html)
- [Multimodal API Input Guide](https://oxalpha.io/blog/ox-alpha-multimodal-guide.html)

## Public website notices

- [Privacy Notice](https://oxalpha.io/privacy.html)
- [Website Terms](https://oxalpha.io/terms.html)
- [Security Information](https://oxalpha.io/security.html)

## Full reference

For a longer, plain-text site summary, see [llms-full.txt](https://oxalpha.io/llms-full.txt).