A new, free artificial intelligence model named Ox Alpha has emerged, capturing the attention of developers and tech leaders in Silicon Valley since its appearance on OpenRouter on Thursday. The model, described as a stealth offering from an anonymous third-party provider, is designed for coding, sustained agentic work, and production workloads, excelling in long-horizon software engineering, complex reasoning, and workflows combining text with visual context.
Ox Alpha is currently available for free with what OpenCode, an open-source AI coding agent, called near unlimited usage for a week. Its provider reportedly has the capacity for 100 trillion tokens per day, a volume approximately 100 times greater than the monthly AI token usage reported by Visa.
Prominent tech figures have already taken notice. Stripe CEO Patrick Collison shared on social media that he found the model very impressive after trying it. Early speculation regarding its origin has pointed towards a Chinese AI lab, with some suggesting Z.ai, the lab behind GLM-5, as a possibility. This theory is supported by observations of similarities in Ox Alpha’s tokenizer behavior and responses to GLM-5, which was previously tested anonymously as Pony Alpha.
This speculation aligns with the growing presence of Chinese AI labs like Zhipu, DeepSeek, and Moonshot AI, which are increasingly challenging US competitors. Their models often approach the performance of leading American systems at a lower cost and are largely open-source. For instance, Moonshot’s Kimi K3, a 2.8 trillion-parameter open-weight model released in July for coding, reasoning, and agentic tasks, quickly gained attention for its performance and affordability.
However, the evidence remains inconclusive. A competing analysis suggests that Ox Alpha’s tokenizer could instead indicate a connection to Microsoft’s MAI family of models. By Saturday, Andrew Curran, an AI analyst, noted that while GLM was the leading theory on Friday night, certainty about the model’s origin had diminished.
