Subquadratic has released SubQ 1.1 Small, a compact AI model with a technical report published on the company's site.
The startup posted the report at subq.ai, marking at least its second model iteration given the 1.1 version number. The "Small" designation follows a naming convention now standard across the industry. The launch drew 31 points and 13 comments on a developer forum at time of publication — modest traction that suggests Subquadratic is still an unfamiliar name outside a narrow slice of the machine learning community.
The small-model segment has become one of the most competitive categories in AI. Meta, Mistral, Google, and Microsoft have all fielded lightweight variants over the past two years, each optimized to run on constrained hardware at lower cost. Fitting into that tier without name recognition is a steep climb, and whether SubQ 1.1 Small has a benchmark story worth telling isn't legible from the top-line announcement alone.
The company's name borrows from computational complexity theory, where "subquadratic" describes algorithms that scale better than O(n²). It's a pointed branding choice — either a genuine architectural claim or a signal aimed at researchers who will recognize the reference.