Category primer

Cheap judgment.Hard decisions.

Decision Models (System One / classifier models) return typed answers with calibrated probabilities — not prose. Use them for the fast gate; keep frontier LLMs for the hard calls.

Live contrast

System 2 · LLM

Thinking…

System 1 · Decision

{}

Orders of magnitude, not vibes

Illustrative for a classification-shaped task — not a vendor benchmark. Real products vary by schema, state size, and hosting.

Latency

Decision Model
~0.2s
Frontier LLM
~8s

Relative cost

Decision Model
Frontier LLM
~200×

Dual-process in one glance

Gate first. Escalate only when the decision is hard or uncertain.

InputSystem OneTyped answerEscalate → System Two

Judgment meter (demo)

Pick a scenario. Bars are a teaching mock — not a live API call.

  • transactional0%
  • marketing0%
  • phishing0%

Go deeper