SubQ

SubQ is a large language model built on a sub-quadratic sparse-attention architecture, offering a 12 million token context window with substantial speed and cost improvements over existing models.