Subquadratic Claims 1,000x AI Efficiency Gain, Researchers Skeptical

Subquadratic, a Miami-based startup that just emerged from stealth, is making bold claims about its new AI model, SubQ. The company asserts that SubQ achieves a staggering 1,000x efficiency gain over existing transformer architectures by escaping the quadratic attention constraint that has long limited model scalability. If true, this would represent a fundamental breakthrough in AI efficiency, potentially enabling much larger and more capable models without proportional increases in compute costs. However, the AI research community has responded with deep skepticism. Several prominent researchers have called for independent verification, noting that such a dramatic improvement would challenge well-established scaling laws. The quadratic attention mechanism is a core bottleneck in transformers, and many teams have tried—and failed—to overcome it without sacrificing quality. Subquadratic has not yet released detailed technical papers or open-sourced its model for peer review. The startup claims its approach uses a novel mathematical formulation that reduces computational complexity without losing expressiveness. But until independent benchmarks are published, the claims remain unproven. If Subquadratic can validate its results, it could reshape the economics of AI development, making advanced models accessible to smaller organizations. For now, the industry watches with cautious interest, waiting for evidence that such a leap is truly possible.