Gemma 4 deep dive: why a 1.5 GB model scores 37.5% on competition mathematics, how the MoE routing actually works, and which model fits your hardware. Full breakdown inside.

The article provides a technical deep dive into Google's Gemma 4 model, explaining how its 1.5 GB size achieves a 37.5% score on competition mathematics through a Mixture-of-Experts (MoE) architecture. It details the mechanics of MoE routing and offers guidance on selecting the appropriate model variant for different hardware, from low-power devices like a Raspberry Pi to high-end research workstations.