Ox Alpha processed 11.6 trillion tokens during its first 72 hours on OpenRouter, roughly 2.6 times the previous record for a model debut on the platform. The model appeared on August 20 without corporate branding and was offered free through August 24. Its 1,048,576-token context window, multimodal support for text, images and video, and maximum output of about 131,000 tokens made it especially attractive to coding agents (automated software-development systems). Early benchmarks recorded success rates above 80% on coding tasks, including evaluations where it outperformed GPT-5.6-sol. No organization has officially claimed Ox Alpha, although similarities in tokenizer behavior and output patterns have prompted speculation about a connection to Z.ai’s GLM model family. Usage remained elevated after the free preview ended, indicating that developers continued to value the model’s performance despite uncertainty over its ownership, data practices, training-data provenance and capacity guarantees.