Subquadratic releases SubQ 1M-Preview, an LLM with subquadratic architecture that reduces attention compute by almost 1,000x compared to other frontier models

Subquadratic, an AI company, launched SubQ 1M-Preview, the first LLM built on a fully subquadratic architecture where compute grows linearly rather than quadratically with context length. The model achieved a research result at 12 million tokens.

SubQ is available via three interfaces: an API for developers and enterprise teams; SubQ Code, a coding agent that loads entire codebases into a single context window via command line interface; and SubQ Search, a long-context search tool providing Deep Research capabilities.

The subquadratic architecture reduces attention compute by almost 1,000x compared to other frontier models, addressing a fundamental limitation of transformer-based LLMs where compute requirements scale quadratically with context length.