Subquadratic AI Breaks Context Window Records

A new AI model called Subquadratic just shattered records by processing a massive 12 million tokens at once. That's like reading a huge book in one go. Here's the thing: It uses a clever new design called Subquadratic Selective Attention. This avoids the slowdowns that plague other AI models when handling long texts. What's interesting is that this new model even beat GPT-5.5 on a key benchmark, scoring 83 compared to 74. It also achieved 92.1% accuracy on a difficult retrieval test with its massive context window. Subquadratic claims its technology is 52 times faster than existing systems for large inputs. They've already launched a beta API and plan an even bigger model with a 50 million token capacity. The bottom line: This could mean much smarter AI that understands and uses far more information than ever before.