Subquadratic is a frontier AI research company developing models and algorithms designed to scale efficiently in compute, memory, and data use. Its flagship model, SubQ, is built for reasoning across multi-million-token contexts, helping enterprises work with large collections of information without relying on chunking or compression. Use cases include analyzing complete software repositories, legal documentation, vendor contracts, and years of financial filings; supporting long-running agents; and retrieving insights across company data.

SubQ uses Subquadratic’s proprietary Subquadratic Sparse Attention (SSA) algorithm, which focuses computation on relevant tokens and relationships. The company states that at a two-million-token context, SubQ uses 128 times less compute than frontier models. Its published benchmarks cover long-context retrieval, graduate-level science, finance automation, and competitive programming. Subquadratic offers access to its models and an API for enterprise projects, alongside research and technical reports explaining its approach. The company’s work focuses on architectural advances in large language models, with the goal of making data-intensive workloads more practical and efficient.