muus research

Independent AI research · UK

Model architecturesfor Apple silicon.

Research shaped by the machine it runs on.

Explore the research
Open weights. Local inference.muus.cc

The architecture begins
with the hardware.

We research model architectures for Apple silicon: open weights, local inference, and execution on physical hardware.

Our benchmark method is batch size one, with stated conditions and reproducible code. The aim is to understand how a model runs on the machine in front of us.

Open weightsBatch size onePhysical silicon
02

muus-conform-1.5B

A ~1.5B model trained from scratch for exact structured output without constrained decoding. Standard architecture; not silicon-specialized.

Phase 0 testbed · 1.5B
In R&D