Subquadratic, a company developing a novel generative AI model, launched today with $29 million in seed funding.
The new large language model, called SubQ, uses a subquadratic architecture that greatly increases the context window-the amount of information the AI can process at once-without a proportional increase in compute power. The company claims it outperforms other state-of-the-art models on speed and accuracy.
Where industry standard models max out at 128,000 tokens, and frontier cloud models like Claude Sonnet 4.7 and Gemini 3.1 Pro reach 1 million tokens, SubQ can handle up to 12 million tokens-equivalent to about 9 million words or nearly 120 books.
To achieve this, co-founders Justin Dangel (CEO) and Alexander Whedon (CTO) engineered a proprietary transformer architecture that replaces dense attention with sparse attention. Traditional transformers compare every token to every other token, creating a quadratic scaling problem where doubling input quadruples compute. Subquadratic's approach achieves linear scaling, so doubling input only doubles compute.
The company reports SubQ is more than 50 times faster and 50 times less expensive than leading frontier models at 1 million tokens while maintaining higher accuracy. At full 12 million-token capacity, compute requirements are reduced by nearly 1,000 times. On the RULER 128K benchmark, SubQ scored 95% accuracy at $8, versus 94% accuracy at $2,600 for Claude Opus-a 300x cost reduction.
Subquadratic is launching the SubQ API for developers and enterprises, and SubQ Code, a coding agent that loads entire codebases into a single context window. The company also plans a search product initially free.
Investors include Javier Villamizar (former SoftBank Vision Fund partner), Justin Mateen (Tinder co-founder), and early investors in Anthropic, OpenAI, Stripe, and Brex.