Subquadratic claims to solve decade-long LLM mathematical bottleneck with SubQ model

According to MIT Technology Review, Miami-based startup Subquadratic claims to have solved a mathematical bottleneck limiting large language models for nearly a decade with its SubQ model, which it says processes up to 12 times more text simultaneously than competing models while matching performance of Google DeepMind, OpenAI, and Anthropic on coding tasks. Initial claims were met with skepticism due to limited independent verification, though the company has begun releasing independent evaluation results.