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
| Model | Current version | Best for |
|---|---|---|
| Haiku | Haiku 4.5 | Fast, cheap, high-volume tasks — quick questions, simple drafts, bulk work. |
| Sonnet | Sonnet 5 (default) | The all-rounder: coding, writing, research, everyday professional work. The right default for most people. |
| Opus | Opus 4.8 | The 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.