Subquadratic releases SubQ 1.1 Small model card
The model achieves near-perfect retrieval up to 12M tokens with 64.5x less compute than dense attention.
See the latest news and media coverage for Subquadratic. We track all announcements, press releases, and industry mentions in real time, all in one place.
Sub-quadratic large language model developer
subq.aiLast updated
In short: Subquadratic launched SubQ, its first fully subquadratic LLM, claiming massive efficiency gains and a 12 million token context window.
The model achieves near-perfect retrieval up to 12M tokens with 64.5x less compute than dense attention.
Speed is 52x faster than Flash Attention. SWE Bench Verified scores 81.8%. Ruler (128K) achieves 95%. MRCR V2 reaches 65.9%.
SSA enables linear scaling attention, achieving 52.2× speedup at 1M tokens and strong benchmarks in retrieval and software engineering.
It enables million-token contexts with frontier accuracy, faster inference, and lower costs. Early access available via API, SubQ Code, and SubQ Search.
Miami-based AI startup Subquadratic came out of stealth mode last month with a huge claim. It announced that it had solved a mathematical bottleneck that...
Miami startup Subquadratic says its SubQ model could make AI 1,000 times more efficient and handle 12 million tokens, but researchers remain divided over whether...
AI thrives on data but feeding it the right data is harder than it seems. As enterprises scale their AI initiatives, they face the challenge...
Track Subquadratic and your other target companies to get real-time alerts and weekly summaries delivered straight to your inbox.
Browse news for competitors to Subquadratic and other trending companies.