The short answer
Claude comes in three sizes: Haiku (fastest, cheapest, simplest tasks), Sonnet (balanced, the default), and Opus (most capable, slowest, most expensive). For most business work, Sonnet is the right choice. Use Haiku for high-volume simple tasks; use Opus for tasks where intelligence matters more than cost.
The three sizes, in plain terms
Anthropic ships Claude in a tiered family. Each model is a trade-off between intelligence, speed, and cost.
Haiku is the small, fast model. It’s good at well-defined tasks: classify this email, extract these fields from a form, summarise this short article. It’s not great at nuanced reasoning, long documents, or complex code. Cheapest by far — perhaps 1/15th the cost of Opus.
Sonnet is the middle model and the default for most work. Smart enough for almost anything you’d hand to a competent generalist. Reads long documents well. Writes good prose, good code, useful analysis. Costs roughly $3 per million input tokens — affordable for serious daily use.
Opus is the biggest model. The smartest, the slowest, the most expensive. You feel the difference on hard problems — multi-step reasoning, ambiguous instructions, tasks where Sonnet starts to wobble. About 5x the cost of Sonnet on the API.
The version numbers (Sonnet 4.5, Opus 4.7, etc.) update over time. The size hierarchy stays the same.
Which to use
A practical rule:
- Start with Sonnet. It’s the safe choice. If your task works on Sonnet, you’re done.
- Drop to Haiku when you’re running the same simple task at high volume and Sonnet feels like overkill. Classification jobs. Bulk extraction. Simple yes/no questions on thousands of items.
- Move up to Opus when Sonnet visibly fails. Complex reasoning across many sources. Subtle drafting where tone matters. Code that has to navigate a large existing codebase.
You should be able to feel when you need Opus — if you’re squinting at Sonnet’s output thinking “this isn’t quite right,” try Opus. If you’re squinting at Sonnet thinking “this is more than I needed,” try Haiku.
What you actually pick in different interfaces
Where you make the model choice depends on which Claude product you’re using.
Claude Chat (claude.ai). A dropdown at the top of the chat lets you pick the model. Pro users have access to all three.
Claude Projects. Same dropdown. The model you pick applies to that project.
Claude API (via n8n or your code). You specify the model in each request. This is where most cost-conscious users mix and match — Haiku for the bulk classification step, Sonnet for the summarisation step, all in the same workflow.
Cowork. Pick when you set up the task. Stick with Sonnet unless you have reason not to.
The mistake to avoid
Two common ones.
Always-Opus syndrome. “Opus is the smartest, so I’ll always use Opus.” Costs add up fast. For most tasks, you’re paying 5x for an output that’s 5% better. Use Sonnet by default.
Always-Haiku syndrome. “Haiku is cheap, so I’ll save money.” On nuanced tasks, Haiku produces output that requires editing — sometimes more editing than just doing the task by hand. The “saved money” was illusory.
The right model is the smallest one that does the job well. Test your task on Haiku first. If it works, great. If it doesn’t, move up.
On the version numbers
Anthropic releases new versions of each model every few months. Each release is broadly better than the one before. As of writing, the current top-of-line is Claude Opus 4.7.
You don’t generally need to track version numbers carefully. Anthropic deprecates old versions on a schedule (typically a year or so after release), and your tooling will tell you when to upgrade. Just use whatever the latest version of your chosen tier is.
Last reviewed: 7 May 2026
Done with the basics? Try a recipe.
Browse Beginner recipes