Subquadratic

Subquadratic is a frontier AI research and infrastructure company that built the first large language model with fully sub-quadratic scaling. Its model SubQ uses Sparse Subquadratic Attention (SSA) architecture — selectively focusing only on token relationships that matter — to achieve linear compute scaling with context length rather than the exponential cost of standard transformers. SubQ supports 12 million token context windows at one-fifth the cost of leading LLMs at 150 tokens per second, enabling use cases like full-repository code reasoning and long-running agent state. The company offers an API for developers and SubQ Code for coding agents, and was built by 11 PhD researchers from Meta, Google, Oxford, Cambridge, and ByteDance.