Kimi K3 Demo Runs a 2.78 Trillion-Parameter Model on One CPU With 8.24 GB of RAM

A public C99 inference project says it can run Moonshot AI's 2.78 trillion-parameter Kimi K3 model on a single CPU with a measured peak memory footprint of 8.24 GB, without a GPU, BLAS library or machine-learning framework. The result does not make a frontier-scale model fast on a laptop, but it reframes the hardware bottleneck: the checkpoint still occupies 1.56 TB on disk, while memory is treated as a speed dial rather than a hard requirement.