SubQ AI: Fast Document Processing & Sparse Attention

A new large language model called SubQ claims it can read vast amounts of documents without slowing down. This could address a significant problem in AI: the high cost of processing very long prompts. SubQ, from the Miami startup Subquadratic, reportedly uses a method called sparse attention. This means it focuses only on the most important relationships between parts of the text, rather than comparing every single piece. Independent benchmarks support these claims, showing strong results in finding facts within large texts and solving coding problems. For example, it achieved 98 percent accuracy at 12 million tokens in one test. Subquadratic also states SubQ uses significantly less computing power and runs much faster than other methods. While these are big claims, the model isn't widely available yet, so the AI community is watching closely. This technology could allow AI to handle entire legal contracts or whole codebases, changing how professionals work with large files.