Subquadratic Launches With $29M, 12M-Token AI

Subquadratic has launched with $29 million in seed funding, introducing SubQ, a large language model featuring a 12 million-token context window — roughly 120 books of text. The company claims its sparse attention architecture makes SubQ over 50 times faster and cheaper than leading frontier models at 1 million tokens, reducing compute costs by nearly 1,000 times at full context length. Unlike traditional transformer models that compare every token to every other token, SubQ uses linear scaling to dramatically cut computational overhead. The company is launching a developer API and SubQ Code, a coding agent that loads entire codebases into a single context window.