SubQ 1.1 Card: Linear-scaling sparse attention with 98% retrieval at 12M tokens [pdf]

(subq.ai) by mitchwainer | Jun 16, 2026 | 0 comments on HN