A mysterious AI called 'Ox Alpha' has appeared and is currently undergoing free testing. The context window is an anonymous lab that can process trillions of tokens per day for 1 million tokens and is providing an API.

OpenRouter and OpenCode are conducting free testing of a mysterious anonymous AI called ' Ox Alpha .' The development organization and detailed specifications of Ox Alpha are unknown, but several users have reported their performance test results.
Ox Alpha - API Pricing & Providers | OpenRouter
OpenRouter offers a service that allows AI development organizations to deploy AI models as ' stealth models ,' concealing their official names and development organizations, in order to provide them with testing opportunities before official release. Ox Alpha is one such stealth model, released on August 21, 2026, and its API is available for free trial for a limited time.
The following is an overview of Ox Alpha published by OpenRouter. The context window has 1 million tokens and is a multimodal model that can take text, images, and videos as input and output text.

Furthermore, OpenRouter statistics show that it handles trillions of tokens worth of traffic daily. This suggests that organizations providing Ox Alpha likely possess large-scale AI infrastructure. Checking the status information published by OpenRouter reveals that while connectivity became unstable around 4:30 PM on August 24, 2026, the service has generally maintained a good performance since its launch.

Ox Alpha has also announced support for the AI agent app Hermes Agent, and Nous Research, the developer of Hermes Agent, states that it is 'capable of processing 1,000 trillion tokens per day.'
Ox Alpha is free via Nous Portal for a limited time. We have capacity for 1 quadrillion tokens per day.
— Nous Research (@NousResearch) August 21, 2026
Let the tokens flow at https://t.co/4KFwhsReyA
Additionally, Ox Alpha is now available for free on OpenCode's cloud-based service, Go .
Ox Alpha is now available on OpenCode Go too
— OpenCode (@opencode) August 21, 2026
For the next 6 days, usage is near unlimited and completely free
It won't count against your Go usage https://t.co/Bq4O6v3xM9
Several performance test results for Ox Alpha have been posted online. The following user reported that 'by creating a fluid simulation application using Ox Alpha and a GPU, better results were obtained than with Qwen3.8 27B and Claude Opus 4.5.'
Ox Alpha one-shotted the GPU accelerated fluid simulation in a single 1000 line html file
— Lisan al Gaib (@scaling01) August 21, 2026
it looks absolutely stunning and is 1000x better than what Qwen3.8 27B or Opus 4.5 did when I tested them earlier this week https://t.co/90brvRzf9m pic.twitter.com/SJame7jWX1
The following user reported that when creating a 3D model of a rocket engine using three.js, Ox Alpha produced inferior results compared to Claude Opus 5, and the output time was longer than that of Claude Opus 5.
Ox Alpha vs Opus 5
— Tim Jayas (@TimJayas) August 21, 2026
i asked both models to build rocket engine using three.js
ox-alpha is not even close to opus 5 performance and it took equal or more time for the output
prompt used for both:
'Create a 3d SpaceX raptor engine using three.js, use white theme background so the… pic.twitter.com/aL739Rm075
It has also been reported that 'while it achieved a higher score than GPT-5.6 Luna when running cybersecurity benchmarks, it did not come close to cutting-edge models such as Grok 4.6 and GLM-5.3.'
OK guys I'm sorry to disappoint but Ox Alpha is overhyped
— pilvar (Philippe Dourassov) (@pilvar222) August 22, 2026
We ran it on our cybersecurity benchmark, it's better than GPT-5.6-Luna, but worse than every other frontier models pic.twitter.com/DtOO18fCBL
The official name, development team, and official release date of Ox Alpha remain completely unknown, and attention is focused on future developments.
Related Posts:
in AI, Posted by log1o_hf







