Five minutes of settings, months of peace
Before you hand Claude real work, it's worth stopping for five minutes on three decisions: which plan, which model, and what Claude remembers and keeps about you.
These aren't technical decisions. They decide how much you'll pay, how fast you'll hit the ceiling, and what happens to your clients' information. You make all three once, and only check them again when something changes.
Which plan to choose
Prices in US dollars, from the official pricing page, as of October 2026, excluding taxes:
| Plan | Price | Who it fits |
|---|---|---|
| No-cost plan | $0 | Getting to know Claude and trying it out |
| Pro | Monthly: $20. Annual: $17 a month ($200 up front) | The starting point I recommend for business owners |
| Max | 5x Pro for $100, or 20x for $200. Monthly only | People who hit the ceiling again and again |
| Team | Standard seat: $25 a month, or $20 on annual. Premium seat: $125 or $100. 2 to 150 seats | A team that works together |
What's missing on the no-cost plan? Claude Code, the Opus and Fable models, Research mode, delegating and scheduling tasks, and Claude in Chrome. The unified Claude from the last lesson will also only reach it "soon". What it does have: conversation, web search, file creation, memory, connectors, Docs, Slides and Design, and up to five projects.
When do you move to Max? Your limit is measured in a rolling five-hour window, and paid plans also have a weekly limit. You can see your usage in Settings > Usage. Hitting the ceiling a few times a week, or working a lot in Claude Code? That's the time. And there's a middle path: when your limit runs out, paid plans let you keep going with usage credits, which means paying per use at API prices.
My recommendation: start on Pro, and move to Max only when you feel the ceiling.
Four models, and how to choose
Inside Claude there are several "brains", in other words models. As of October 2026, these are the four current ones:
- The strongest: Fable 5.1. For demanding thinking and long, complex work, and the slowest of the four. Watch the cost: on Pro it isn't included in your limit, and it's paid from usage credits from the very first message. On Max and on a Team Premium seat you can use it for up to half of your weekly limit. On the no-cost plan it isn't available.
- The workhorse: Opus 5.5. For complex work and long tasks. In the developer docs, this is the model Anthropic recommends starting with when you're not sure.
- Fast and smart: Sonnet 5.5. According to Anthropic, the best mix of speed and intelligence. Great for everyday tasks where the path is clear.
- Small and cheap: Haiku 5.5. The fastest, for simple tasks that repeat a lot.
My rule: think on a strong model, execute on a fast one. I run the planning stage of every serious task on Opus 5.5. I move to Fable when something is truly complicated, and on Pro I remember that it costs extra. Drafting and tidying go to Sonnet, and simple background jobs go to Haiku.
Where do you switch? In the model menu next to the send button. That's also where you choose the effort level, from Low to Max, and the recommended level for each model is marked "Default". Higher effort gives more thinking, but it also burns through your limit faster.
What Claude remembers, and what's kept about you
Memory
Since August 25, 2026, memory is on by default on the no-cost, Pro and Max plans, and it carries across conversations and tasks. On Team and Enterprise it's off until someone turns it on. This is very convenient: Claude remembers who you are, what your business is and how you like to work. And that's exactly why you should check what it remembers.
- See and edit: in
Settings > Your Claude, click "Manage" next to "Your memory". Everything there is organized by topic (Topics), and you can edit or delete any item. In the old interface it's inSettings > Memory. - Sensitive topics: health, beliefs and the like aren't saved unless you've turned on "Include sensitive topics in memory".
- Turn it off: the "Memory from chats" toggle on the same screen. You can pause it, so the memory is kept but not used, or reset it, so everything is deleted, including project memory, with no way back.
- For one conversation: in a new conversation, in the "+" menu, turn off "Memory" before the first message. And for a conversation that isn't saved at all, there's incognito chat, under the ghost icon.
The training question
The question I hear at every talk I give: "Is Claude training on my data?" On the no-cost, Pro and Max plans it comes down to one toggle: in Settings > Privacy, under "Help Improve our AI models".
When the toggle is on, new conversations may be used for training, and kept separate from your identity for up to five years. When it's off, new and old conversations won't be used for future training, and only what already went into a training run that has started stays there. Incognito conversations aren't used for training either way. All the details are in the official privacy center.
My position for a business that works with client materials: off. On Team and Enterprise, according to the pricing page, there's no training on your content by default. And here's what doesn't depend on any toggle: passwords, credit card details and clients' medical documents don't get pasted into a conversation. Claude's help center also recommends being careful with exactly this kind of information.
Learn more, at no cost
Anthropic has an official learning site, Claude Academy, with courses in English at no cost (same name as mine, different academy). You sign in with a regular Claude account, even on the no-cost plan, and your progress is saved. What I recommend there:
- For small businesses: the course "AI Fluency for small businesses", eight lessons and three and a half hours, which Anthropic built together with PayPal and two small businesses.
- For the basics: "Claude 101" and "Introduction to Claude Cowork".
- If you're on your way to building: "Claude Code 101".
There's also a collection of short guides for small businesses there, "Claude for small business". And the courses are also on the older site, anthropic.skilljar.com, where sign-up is also at no cost.
Try it now
- Go to
Settings > Usageand check how much of your limit you've used this week. - Go to
Settings > Privacyand decide, on purpose, where the "Help Improve our AI models" toggle should be. - Go to
Settings > Your Claudeand make sure memory is set the way you chose. - Open a new conversation and send:
"What do you remember about me and my business? Give me a tidy list organized by topic, and flag every item that looks out of date or inaccurate so I can fix it."
- Fix whatever's wrong directly on the memory screen.
What to take from this
These settings decide how much you'll pay and what happens to your clients' information, so choose them on purpose. Start on Pro, think with a strong model and execute with a fast one, and check every so often what Claude remembers about you.
