Ox Alpha processed 11.6T tokens in three days, 2.6x OpenRouter's previous biggest model launch.

This kind of scale is only possible now, because of coding agents, where long contexts, retries, and repeated tool calls can massively multiply token consumption inside one task.

Ox Alpha has a 1.05M-token context window and is intended for sustained agentic work, so each session can carry far more text than ordinary chat.