Subquadratic SubQ 12M-Token Subquadratic LLM
Key Questions
What is SubQ?
SubQ is a sub-quadratic large language model launched by a $29M seed-funded startup in Miami. It uses sparse attention to enable a 12M-token context window, achieving state-of-the-art performance in long-horizon reasoning without quadratic compute scaling.
What makes SubQ's architecture special?
SubQ features a fully subquadratic architecture where compute grows linearly rather than quadratically with context length. This allows faster and cheaper inference, positioning it to challenge models like DeepSeek and GPT-5.5.
What is the current status and reception of SubQ?
SubQ is in development, with the first model being SubQ 1M-Preview. It has generated significant buzz on Hacker News, signaling a potential efficiency breakthrough for AI agents, reinforced by multiple media coverages.
$29M seed Miami startup launches SubQ with sub-quadratic sparse attention enabling 12M ctx SOTA long-horizon/reasoning without quadratic blowup, faster/cheaper inference challenging DeepSeek/GPT-5.5; HN buzz signals efficiency breakthrough for agents; multiple coverage reinforces momentum.