Two nanos, a Haiku, and an occasional Sonnet: the roster is the price ladder — 1× / 3.1× / 12.5× / 25× on output tokens. The interesting question is never "which model is smartest" — it's "when does the 25× model earn 25×." For ordinary tasks, cheaper usually suffices; the premium lane earns its price on the hard ones. Watch letter-count, where it was the only lane that could count — and email-hunt, where it was right and still lost to a nano.
Prices as listed by OpenAI and Anthropic, July 2026.