Model families
Four names sit in one menu, and for most jobs one of them is already the right answer. Here is what each is for, and when moving is worth it.
“Haiku for fast cheap answers. Use Sonnet for most things. Opus when you want it to review or orchestrate. Fable is the most powerful but burns credits, so avoid it generally.” My rule of thumb, not Anthropic’s.
Version names as at 17 September 2026: Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5. The families keep their names; the numbers move often, so check Anthropic’s models overview before trusting a version.
This page is about the menu next to the send button. What Claude is made of, block by block, is Claude 101 →
Haiku
The fast one, for answers you want now
Anthropic call Haiku 4.5 the fastest model with near-frontier intelligence.
Reach for it when speed matters more than depth.
- Is this a quick lookup, or a real piece of work?
- Do you need it now more than you need it perfect?
- Would you check the answer yourself anyway?
In Cowork on a paid plan it holds less of one conversation than Fable 5.1, Opus 5 or Sonnet 5, and it has no effort setting of its own.
Sonnet
The default, and where most of your work belongs
Anthropic describe Sonnet 5 as the best combination of speed and intelligence.
When you chat with Claude on a paid plan, it holds as much in one conversation as Opus 5 or Fable 5.1.
- Is there a reason this job needs more than Sonnet?
- Have you tried it here first?
- Is the prompt the problem, not the model?
Start here and stay here. In lessons, the people who run out of usage are usually on a heavier model than the job needed.
Opus
Bring it in for the hard or the high-stakes
Anthropic build Opus 5 for complex agentic coding and enterprise work.
It is the one to hand a job that has to be right.
- Is this high-stakes, or many steps long?
- Do you want it to check what Sonnet just made?
- Is it worth the extra usage?
Anthropic’s developer docs say most workloads start on Opus 5. That is written for developers building on the API, not for a plan with weekly limits.
Fable
The most capable, and the one that costs you
Anthropic call Fable 5.1 their most capable widely released model, built for demanding reasoning and agent sessions that run for hours.
It is available on paid plans only.
- Is this genuinely the hardest thing you will ask this week?
- Does your plan include it, or does it bill separately?
- Could Opus do this instead?
On Max it draws on your weekly limits and uses them faster than other Claude models. On Pro it runs on usage credits instead.
Change it in four clicks
It sits next to the send button, and you can change your mind halfway through a chat.
- 1
Look next to the send button. The model and the effort level are both shown there.
- 2
Click the model name. Pick another one, or click More models for more options.
- 3
Set the effort while you are there. Low and Medium stretch your usage further.
- 4
Change it mid-chat if you need to. It applies from Claude’s next response, not backwards.
On an Enterprise plan, a model or effort level you expect may be missing because your administrator turned it off for your role.
What that menu actually controls
One menu, three settings. Most people change the first and never touch the other two.
Model
which Claude answers- Four current names.
- The one most people change.
Effort
how hard it works- High is the best overall balance.
- Lower stretches your usage.
Thinking
reasoning you can open- A section above the answer.
- Cannot be turned off on Fable 5.1 or Opus 5.
Anthropic’s developer docs: tuning effort is often a better lever than switching models.
- Haiku: the fastest, for quick answers.
- Sonnet: the default, and where most work belongs.
- Opus: for the hard, the high-stakes, and the review pass.
- Fable: the most capable, and the one that eats your limits.
This week: leave it on Sonnet, and only move when you can say out loud why.
What all this costs you, and the plan questions: FAQ → The same models in a different room: Chat vs Cowork →
Do it with someone. Everything here is free and you can run it alone. If you would rather work through it with me, one to one, book a free 30-minute chat. No pitch.