Subquadratic

Subquadratic is an AI infrastructure and research company pushing the boundaries of what's possible with large language models. Unlike companies focused on incremental transformer improvements, Subquadratic is building a new class of LLMs on a proprietary post-transformer architecture that enables large-context, multimodal inference scaling efficiently where traditional transformers struggle.

The team believes efficiency is intelligence - and they're putting that philosophy into practice by developing alternative model architectures that can handle longer contexts and more complex tasks with better computational efficiency. Researchers and engineers at Subquadratic regularly participate in top AI conferences like NeurIPS, AAAI, and ICLR, contributing to the broader research community while hiring talent interested in novel LLM architectures, large-scale model training, and multimodal systems.

The company offers developer-facing products including a Speech-To-Text API, with a focus on building robust infrastructure that serves both real-time streaming and batch processing needs. Their Miami-based team spans research, engineering, and go-to-market functions.