What is Ox Alpha on OpenRouter?

Ox Alpha is an anonymous LLM offered through OpenRouter’s Stealth provider. It is designed primarily for software engineering and agentic coding, supports a 131,072-token context window, and is currently free to use. OpenRouter has not identified the company or research lab responsible for developing the model.

How many tokens has Ox Alpha processed?

OpenRouter’s leaderboard shows 17.5 trillion Ox Alpha tokens for the week through August 24, 2026. On August 24, OpenRouter said the model was approaching 6 trillion tokens for that day alone. An earlier snapshot reported 11.6 trillion tokens during its first two days of availability.

Is Ox Alpha made by Z.ai?

Z.ai has not claimed Ox Alpha, so its involvement remains unconfirmed. Community testers have identified behavioral similarities with Z.ai’s GLM models, including coding style and apparent self-identification in some responses. Those fingerprints make GLM a plausible theory, but model-generated identity claims and output similarities are not conclusive evidence.

Why is Ox Alpha called Niu Lai?

Chinese internet users call Ox Alpha “Niu Lai” after Here Comes the Ox, a mother-and-son animated yak movie that became a meme. The shared ox reference made the film an easy nickname for the anonymous model. Some users also call Ox Alpha the “Bull Demon King,” but neither name identifies its developer.

Is Ox Alpha an open-weight model?

Ox Alpha is not currently an announced open-weight model. It is free through OpenRouter, but its weights, architecture, training details, and license have not been published. Free API access allows developers to use a hosted model without token charges; open-weight access would allow them to download and run the model under a stated license.

AI News08/25/26August 25, 20268 min

Jim Clyde Monge

Ox Alpha Nears 6T Tokens in a Day on OpenRouter

The free coding LLM has dominated OpenRouter’s rankings, while a cow-movie meme and GLM fingerprints fuel the hunt for its maker.

Last updated on August 25, 2026

AI Summary

Ox Alpha approached 6 trillion tokens on August 24, 2026, cementing the anonymous free coding model’s dominance of OpenRouter despite its maker remaining unknown. OpenRouter recorded 17.5 trillion Ox Alpha tokens for the week through August 24, up from 11.6 trillion in its first two days and more than 13 times MiniMax M2.5’s 1.33 trillion. The software-engineering model offers a 131,072-token context window with no input or output token charges, encouraging high-volume coding-agent workflows. Community fingerprints suggest Z.ai’s GLM family, but no lab has claimed it and self-identification is inconclusive. OpenRouter also says prompts and completions may be retained, though not used for training. The traffic reflects consumption rather than superiority, revenue, or user count, and Ox Alpha has no published benchmarks, weights, architecture, license, or long-term pricing. The key tests will come after its stealth period: whether independent evaluations validate its coding performance and whether demand persists once normal pricing arrives.

Ox Alpha’s token count has moved from unusual to industrial-scale. In an August 24 post, OpenRouter said the anonymous model was on track to process nearly 6 trillion tokens that day. The platform’s latest completed leaderboard, updated through August 24, shows 17.5 trillion tokens over the previous week.

The numbers align with the model’s rapid ascent. Business Insider had reported 11.6 trillion tokens during Ox Alpha’s first two days. Adding another 5.9 trillion brings that earlier snapshot close to OpenRouter’s current 17.5 trillion total.

Its maker remains unconfirmed. Community fingerprints lean toward Z.ai and the GLM model family, but nobody has produced conclusive evidence, and no AI lab has publicly claimed Ox Alpha.

Ox Alpha Is on Track for a Six-Trillion-Token Day

OpenRouter’s “today” referred to Monday, August 24, rather than Tuesday, August 25. The completed leaderboard now gives that projection more weight: the difference between the earlier 11.6 trillion figure and the newer 17.5 trillion total is approximately 5.9 trillion tokens.

Spread evenly over 24 hours, 6 trillion tokens works out to roughly 69 million tokens per second across all concurrent requests. It does not mean one instance of Ox Alpha was generating at that speed. OpenRouter aggregates traffic from many applications, developers, coding agents, and automated processes using the model simultaneously.

Ox Alpha’s lead is still extraordinary within that context. Its 17.5 trillion tokens are more than 13 times the 1.33 trillion recorded by second-place MiniMax M2.5. The anonymous model also processed about three times as many tokens as the other nine models in OpenRouter’s top 10 combined.

Raw token volume can rise quickly in agentic software development. A coding agent may repeatedly submit repository files, conversation history, terminal output, test results, and tool responses. One programming task can therefore produce dozens of calls and consume far more tokens than a conventional chatbot conversation.

Free Access Helps Explain the Usage Explosion

The Ox Alpha model card describes it as a text model optimized for software engineering and agentic coding. It has a 131,072-token context window, accepts text input, produces text output, and is currently offered without an input or output token charge. OpenRouter lists its creation date as August 21, 2026.

That combination is built to attract heavy experimentation. Developers can connect a coding harness, give the agent a large codebase, and allow it to run multi-step tasks without watching a metered API bill increase after every attempt. OpenRouter’s post specifically promoted using Ox Alpha through its command-line tool with existing coding agents.