Subquadratic launches SubQ #subquadratic #subq #longcontextai

Subquadratic introduced SubQ, an AI model built for extremely long context with a claimed 12 million token context window. The launch targets one of the biggest limitations in current models like GPT, Claude, and Gemini: the rising cost and slowdown of handling large context through traditional attention. SubQ uses SubQuadratic Sparse Attention, or SSA, to concentrate compute on the most relevant token relationships instead of processing every possible pair. That makes tasks like searching huge codebases, legal archives, research libraries, or thousands of documents more practical at scale. Subquadratic claims SubQ delivers dramatically faster and cheaper long-context performance, including 52x faster processing than FlashAttention at 1 million tokens and less than 5% of the cost of current frontier models. Early access is being opened for SubQ, SubQ Code, and SubQ Search, signaling a broader push toward AI systems designed for real enterprise-scale retrieval, coding, and document analysis. #Subquadratic #SubQ #SubQCode #SubQSearch #SSA #SubQuadraticSparseAttention #LongContextAI #12MillionTokens #FlashAttention #GPT #ClaudeAI #GeminiAI #CodeAnalysisAI #LegalTechAI #EnterpriseAI