Zhipu AI, also known as Z.AI Co., confirmed that Ox Alpha is a new GLM-series multimodal model and said it will publish the model weights tonight. Ox Alpha first appeared anonymously on OpenRouter and OpenCode on August 20, quickly topping global usage charts and OpenCode reported a capacity claim of 100 trillion tokens per day while independent reports said the deployment processed trillions of tokens and broke OpenRouter launch records. The model supports text, image, and video input, shows strong coding and agent capabilities, and offers a million-token context window according to multiple posts and Gateway listings. An independent investigator matched Ox Alpha to a Z.AI GLM-family model by comparing tokenizer counts and an error code after roughly 250 probing attempts. Coverage from Bloomberg and TechCrunch framed Ox Alpha as rivaling other frontier models such as DeepSeek while also noting it launched without a paper, lab disclosure, or compute provenance. Zhipu’s pledge to release weights will enable researchers to audit training lineage, run local instances, and prompt immediate community replication and scrutiny.
