Claude AI - which plan to use?

AI
1

(Following up on another thread about folks switching to Claude from ChatGPT)

In looking at the Claude pricing plan and features, the interesting differences for me seem to be:

Paid $17/month plan adds Claude Code and MCP (two things I would like to experiment with)

Pro plan at $100+/month adds “Memory across conversations”

This has me confused. In my current ChatGPT Plus Plan ($20/month) I have “Expanded memory and context”

Would I really be losing this with Claude?

While using ChatGPT over the past few months, I have been impressed with how it tailors the responses to what I have already been working on in previous sessions, and often asks me follow-up questions that are related to the work I am doing and truly remind me of additional details that I would not always have thought of myself.

Since the features in the plan descriptions can be vague, I can’t really be sure that what I think I am getting with ChatGPT is or is not included in Claude or even is the same thing?

Looking for clarification from others that have been using or have compared both.

3 Likes
2

Claude will use context from whatever chat you’re in, but not context from other saved chats.

Edit
This has only been problematic once when my chat reached its maximum size and I needed to start a new chat. But to be fair I was working on three large documents that Claude had created that amounted to 100 pages of text. I’ve not had this problem since. You can of course copy necessary contextual info from one chat to another.

3

They’ve just rolled out memory to all paid accounts in the last few day. It works a little differently to ChatGPT, but I think you’d be best to google that one to figure out how - I can’t explain it!

1 Like
4

Thanks for that. Here’s the Anthropic post:

October 23, 2025

Claude’s memory on Max and Pro plans

Max plan users can now access Claude’s memory capabilities. This feature will roll out to Pro plans gradually over the next two weeks.

And

You can now prompt Claude to search through your previous conversations to find and reference relevant information in new chats. Additionally, Claude can remember context from previous chats, creating continuity across your conversations. This article introduces Claude’s chat search and memory capabilities and explains how they work, what Claude can and can’t remember, and how you can toggle the features on/off.

This is from their support website.

5

As shared above, memory has now been added. Claude has projects and artifacts features, which I find extremely helpful. In addition to the memory function, you can now keep all related material and chats in a project, which helps Claude focus on project conversations and materials, which in my experience results in better output. Here is a description of those features, per Claude.

Projects and Artifacts: Functional Distinctions

Projects and Artifacts serve related but separate purposes in workflow organization within Claude AI.

Projects are organizational containers. They hold multiple conversations, files, and shared context within a defined scope. If a user is working on a manuscript, for example, they might create a project called “Leadership Book” and conduct all related conversations there—draft reviews, chapter outlines, research summaries, citation checks. The project retains memory across sessions, allowing continuity without repetition. It functions as a dedicated workspace where all materials and dialogue remain accessible and contextually linked.

Artifacts are discrete outputs generated within a conversation. When Claude produces a document, a speech, a report, or a draft chapter, that content appears as an artifact—a standalone file the user can view, download, or edit. Artifacts are the deliverables. They exist independently of the conversation thread and can be saved in various formats: Markdown, PDF, DOCX, and others. Users interact with the artifact directly, not through dialogue.

In practical terms: A project organizes the work. An artifact is the work itself.

A user might create a project for “Board Reports 2025” and within that project generate multiple artifacts—each report as its own file. The project ensures that all discussions about governance, institutional priorities, and strategic language remain in one place. The artifacts are what the user exports to other applications, distributes to colleagues, or archives in cloud storage.

The difference is structural. Projects provide continuity and context. Artifacts provide output and portability. Both serve organizational needs and version control, but they operate at different levels of the workflow.​​​​​​​​​​​​​​​​

1 Like
6

I don’t understand how “memory” in this sense works with confidentiality.

Before “memory”, it seemed like most confidentiality concerns even for legal matters were accommodated by precluding content from being used for training. That would no longer seem applicable.

7

In addition, each project has its on instructions i.e., how you want claude to respond, what personality & capabilities it should have etc (similar to claude.md file for claude code). This is one of the more important aspects of grouping as it can entirely change its efficacy as your intern.

2 Likes
8

FWIW, memory is different from training. Anthropic provides an opt-in option if you’d like your chats to be used for training. Enabling Memory, however, causes Claude to consult past chats when responding to a new prompt.

3 Likes
9

Yes, this is one of the most difficult things for people to wrap their minds around. There is almost nothing you are doing with AI that is “training” it. All of these consumer features like memory, or custom GPTs, or lengthy conversations are simply expanding the context window. They are not actually training any model you use.*

*If you don’t “opt out,” all of your conversation history will be aggregated with millions and millions of other conversations to train future models, no different than updates to Wikipedia or new news articles will be used to train. * *

** Even this is a bit of an oversimplification but cut me some slack!

2 Likes
10

Thank you both, Jacobio and Jezmund_Berserker.

I don’t think I can claim to understand but at least feel somewhat re-assured. Between all my regulator’s warnings about the risks of breaching confidentiality and the opacity of the terms and conditions, it seems very confusing.

1 Like
11

Why even use plan? I’m paying for tokens and never go above $20 in 3 months.

12

How is it that you’re using Claude to use tokens? I naturally have a follow up of how you do it that way, assuming it’s something that would work for me.

13

See: Paying for Claude App vs Paying for Tokens – key apps BoltAI (Chat Interface for many many moldels via SetApp), Zed (an IDE) and ObsidianCopilot (Chat with your notes) + DevonThink.

1 Like
14

depends on use case really. Occasional use is likely better to use pay-by-token. For me, I’d run out of 20 bucks in a day or two (been there). I did the 20/month for a while but started running out in about half a day, then you get 3-4 hours delay before you can pick up again. this worked for me for a while but I am on the 100/month plan now.

15

I’m also on the $100/mo plan and I get far more value out of it than that.

Also, since this thread started, Claude has added the ability to “compact conversation history,” so you can continue those extra long conversations without having to start a new one.

16

If you’re looking for a reason to subscribe to the Claude Pro plan ($20 per month or $204 annually), the Claude Code interface in the Claude desktop app may be it. You can, of course, use the Claude Code CLI in Terminal using an API key, and if you’re comfortable in Terminal, it’s fine.

However, if you’d prefer a non-CLI wrapper around Claude Code, Anthropic now provides one via the Claude desktop app. Claude Code is a powerful tool that you can use for many things other than coding, and I think the new desktop implementation will make it more straightforward for many of us to access them.

Ethan Mollick has a new Substack article up that does a nice job describing what Claude Code is, what it does and how, and a few basics about the desktop version.

Claude Code and What Comes Next: With the right tools, AI can accomplish impressive things

There’s a ton of content out there about all the things you can do with Claude Code besides code. This article focuses on some uses that will likely appeal to at least some of us on this forum:

Claude Code Is Not a Coding Tool—It’s a Personal Assistant That Changes Everything: How to use Claude Code for research, documentation, and automation—with real examples

1 Like
17

One of the reasons to use CC in CLI is that you can have CC use all sorts of bash/zsh commands i.e., it integrates much more deeply into your entire toolset. They may not be important to non-coding use of CC, though even with (say) a text editor app where you are busy writing an article CC can directly obtain the text to grammar/spell check and update the text without having to copy paste. (just a minuscule example of integrating in the CLI)

1 Like
18

True, though the compacted context is not available in all its detail in your current state. Best to let Claude write .md file(s) along the way for what you consider important findings…

19

Understood! But if you’re doing straightforward, non-coding things and you’re not comfortable with the CLI, CC in the desktop app will likely be able to do what you need, and, more importantly, keep you out of trouble.

20

Nice articles. That was a good read.
Just opened Visual Studio, pointed it to my Obsidian folder, installed the Roo Code extension, and entered an OpenRouter API key.
Will ask what it can do.

Interesting times ahead