Claude Models Explained: Which One to Use

Claude comes in three model families — Haiku, Sonnet, and Opus. Here's what each is best at and how to pick, in plain English.

When you use Claude, you're using a specific model — a version tuned for a different balance of speed, cost, and raw capability. As of 2026 there are three families: Haiku (fastest), Sonnet (the everyday default), and Opus (the most powerful). In the claude.ai app you pick the model from a dropdown near the message box.

The Three Families

ModelCurrent versionBest for
HaikuHaiku 4.5Fast, cheap, high-volume tasks — quick questions, simple drafts, bulk work.
SonnetSonnet 5 (default)The all-rounder: coding, writing, research, everyday professional work. The right default for most people.
OpusOpus 4.8The hardest problems: multi-file coding, deep math and reasoning, work where maximum accuracy beats speed.

How to Choose

  • Just starting out? Stay on the default (Sonnet). It handles almost everything well.
  • Need speed or you're asking lots of small things? Haiku is faster and lighter.
  • Stuck on something genuinely hard — a tricky proof, a big refactor, dense analysis? Switch to Opus for that task, then switch back.

A rough rule: Sonnet is the workhorse, Opus is the specialist you call in when quality matters more than speed, and Haiku is the sprinter. On the API, Opus 4.8 costs $5/$25 per million input/output tokens versus Sonnet's $3/$15 — but in the claude.ai app you don't pay per token, you just pick from the models your plan includes. See plans and pricing for which models each plan unlocks.