Subquadratic ✓

SubQ is a sub-quadratic language learning model, the first of its kind, developed by Subquadratic Inc. It is designed specifically for long-context tasks.

The model's primary feature is its ability to reason across a vast amount of tokens - up to 12 million - without compromising on the quality of output.

This capability allows SubQ to work across extensive repositories, long histories, and persistent states, a functionality that may satisfy needs of operations dealing with large data sets and lengthy procedures, such as processing complete Python source code libraries or months of PRs.

\ SubQ revolutionizes the traditional language learning models architecture by implementing a fully sub-quadratic sparse-attention design. Most traditional models process all possible relationships between words, which tends to waste computational resources as only a fraction of these relationships are relevant.

SubQ, on the other hand, only focuses on the connections that matter, ensuring efficient usage of computational resources. \ SubQ also boasts about its pricing being significantly more affordable than other leading language learning models.

For developers and teams, the product offers an API for processing entire repositories and pipeline states in a single API call at a linear cost. For coding agents like Claude Code, Codex, and Cursor, it provides a long-context layer to map codebases, gather context, and answer token-heavy questions faster.

\ Subquadratic Inc. is a frontier AI research and infrastructure company known for challenging the traditional transformer model designs and pushing for a foundational change at the model architecture level to enable large-context, multi-modal inference that scales efficiently.