bestmodel.run — what do you want to run?
An honest compatibility engine for local AI, built from community pool measurements.

Honesty ladder: measured > reported > extrapolated > formula > no data yet.
A cell without a source class never renders. Numbers are never rounded in the flattering direction.

Pool snapshot 2026-09-18:
  runs    8,230
  models  686
  rigs    184

Answers the pool can give (human view renders them interactively):
  can it run?      /hardware — feasibility from bandwidth and VRAM rules
  how fast?        /wall — every community cell with its basis and n
  is it worth it?  /track-record — trust earned by verified acts

Machine surfaces:
  agent twins      append ?as=agent to any route
  contract         /llms.txt
  CLI quickstart   docs/agent-quickstart.md in the repo (git clone, no key)