Why linear complexity matters

Transformers changed everything. But their O(N²) attention mechanism means that as sequences get longer, cost grows fast. For researchers and users without access to large GPU clusters, this is a real barrier.

We are exploring sequence architectures with linear complexity. Models that scale with input length, not against it. The goal is not to replace Transformers everywhere but to find the right architecture for the constraints we care about: consumer hardware, long sequences, and multimodal inputs.

This is early work. We will share what we learn as it takes shape.