Wake up solvers!
A new challenge just dropped!
Last week the community made Laguna XS 2.1 run 2.6x faster on Apple Silicon. Today the target is Qwen3.8 27B from
@QwenDevs, a step-function jump for small local models, and the one everyone insists Macs are bad at.
Dense models are bandwidth-bound.
That's the whole reason "Macs are slow on dense models" became conventional wisdom.
We gave the community early access yesterday.
It's already at 2.5x!