Which Claude
to use
There is more than one Claude. On claude.ai, you will see a model dropdown. This page is how to pick without thinking about it too hard.
Claude comes in a small family of models with different strengths. You do not need to understand why, but you do benefit from picking the right one for the task at hand. The choice takes about three seconds once you know the rule of thumb.
Start with Sonnet. Switch to Opus when a task is hard enough to be worth the wait. Switch to Haiku only if you need something fast and disposable.
The three to know
Claude Sonnet
The one you should use for almost everything.
Sonnet is the workhorse. Fast enough for everyday use, smart enough for almost any real task, and the one you should reach for unless you have a specific reason not to. If you open claude.ai and do not touch the model dropdown, you are probably already using it.
Best for
- Writing first drafts, summarising documents, drafting emails
- Most coding tasks, code reviews, debugging
- Reformatting data, cleaning exports, matching templates
- Meeting notes, status updates, commentary around numbers
- The vast majority of what's on the role pages
Claude Opus
Smarter, slower, more expensive. Reach for it when the task is genuinely hard.
Opus is the deeper thinker. It is noticeably slower than Sonnet and uses more of your usage allowance, but on genuinely difficult tasks it produces meaningfully better results. The test for whether to use it: would a smart colleague need to go away and think for twenty minutes on this? If yes, Opus. If no, Sonnet.
Best for
- Strategy memos, investment memos, board narratives where nuance matters
- Synthesising conflicting inputs from multiple sources
- Complex code problems that Sonnet has already had a go at and got wrong
- Anything where you would be embarrassed to send the first draft
- Long-form writing where voice needs to be consistent over several thousand words
Claude Haiku
Small, fast, cheap. Almost nobody on claude.ai needs to pick this.
Haiku is the fast one. On claude.ai it rarely makes sense to pick it because Sonnet is already fast enough for chat-based work. Haiku matters most when you are calling the API in volume, where the cost and latency difference per call adds up across thousands of requests.
Best for
- Classifying or tagging many small inputs via the API
- Extracting structured data from predictable sources at volume
- Anywhere you need "good enough, a thousand times an hour"
- Almost never what you want in a chat window. Stick with Sonnet there.
The five-second decision
If you do not want to think about this every time you open Claude, use this rule:
- Default to Sonnet. It is the right answer nine times out of ten.
- Switch to Opus when Sonnet gives you an answer that is almost right but not quite, and you suspect the problem is hard rather than under-specified. Try adding context first. If that does not fix it, upgrade the model.
- Do not touch Haiku on claude.ai. It lives in the API world. If you are not writing code that calls Claude programmatically, you can forget it exists.
That is the whole page. There are version numbers (Sonnet 4.6, Opus 4.6, and so on) but those change faster than this handbook can track, and the practical advice does not: pick Sonnet unless you have a reason to go bigger.
One thing worth knowing about switching mid-chat
If you switch model in the middle of a conversation, the new model reads the full conversation so far. It does not inherit the previous model's reasoning, only the text. This is usually fine and sometimes helpful. If Sonnet got stuck on something, switching to Opus and asking it to try again often unblocks you, because Opus gets the same context but brings more to it.
When the output is almost right but not quite, and the prompt is already good, upgrading the model is the next thing to try.
Next: When not to use Claude is the honest counterpart to this page. Some tasks are better done without any Claude at all.