Subquadratic

About Subquadratic

Subquadratic is a frontier AI research company that has developed a new class of large language models (LLMs) built on a fully subquadratic architecture. The company's core innovation addresses the quadratic scaling problem inherent in traditional transformer models, where compute requirements grow quadratically with the length of the input data. Subquadratic's approach allows for compute to grow linearly with the context length, enabling significantly larger context windows, faster inference, and lower costs. The company was co-founded by CEO Justin Dangel and CTO Alexander Whedon. Dangel is a five-time founder with a background in health tech, insurancetech, and consumer goods. Whedon previously worked as a software engineer at Meta and was the Head of Generative AI at TribeAI. In May 2026, Subquadratic announced its launch from stealth with $29 million in seed funding. Investors include Justin Mateen (co-founder of Tinder), Javier Villamizar, and early investors in major tech companies like Anthropic, OpenAI, and Stripe. Subquadratic's first model, SubQ 1M-Preview, is designed to handle extremely long contexts, with research results demonstrating capabilities up to 12 million tokens, which is equivalent to roughly 9 million words or 120 books. This breakthrough in context window size is achieved through a proprietary architecture using Subquadratic Sparse Attention (SSA), which intelligently focuses compute on the most relevant relationships between tokens instead of comparing every token to every other token. This efficiency gain is substantial, with the company claiming its model reduces attention compute by nearly 1,000x at 12 million tokens compared to other frontier models. For its business model, Subquadratic offers its technology to developers and enterprise teams through several products currently in private beta. The "SubQ API" provides full-context access for processing large repositories and data pipelines in a single call. "SubQ Code" is a command-line interface agent that can load entire codebases into a single context window for development tasks. Additionally, "SubQ Search" is a long-context search tool for deep research. The company's models are positioned to be significantly more cost-effective, reportedly being over 50 times cheaper than leading models at 1 million tokens while maintaining high accuracy. Keywords: Subquadratic AI, large language models, LLM, subquadratic architecture, sparse attention, long context window, artificial intelligence, Justin Dangel, Alexander Whedon, SubQ 1M-Preview, AI efficiency, machine learning, natural language processing, AI development, API, coding agent, AI research, compute scaling, linear scaling, transformer models, AI infrastructure, enterprise AI, AI tools, data processing, generative AI