@SubQ

SubQ, a new large language model built on a subquadratic sparse attention architecture, has entered private beta with a 12-million-token context window. The architecture allows compute to scale linear…