Microscale
0
← back to the atlas
Act I · Region 01

The Landscape

Why small models exist — five drivers, zero mythology

Before the first equation we build intuition: when a 70B model serves a billion requests, the deployment curve starts to hurt. Here we unpack the five drivers that pushed the frontier back toward small models.

badge · Cartographer
0 of 4 lessons completed
  1. 1
    Tokens and probabilities
    What a language model really outputs
  2. 2
    The cost curve
    Drag a slider from 1B to 500B and feel the curve
  3. 3
    The five drivers
    Inference · edge · latency · specialization · velocity
  4. 4
    When an SLM fits
    Drag 12 scenarios into the right buckets