I pay for two AI tools right now.
One is Claude. I’m on the $100/month Max plan, and it does somewhere around 90 to 95 percent of my work. The other is Grok. I’ve had a SuperGrok subscription for a while, and I use it for completely different things.
I don’t pay for ChatGPT. I don’t pay for Gemini separately either (I get it bundled with my Google Workspace account). I’ve used them both. I know what they do. They’re just not where I live day to day.
So, I’m going to be upfront about that. I’m a big Claude fan. But this post isn’t me telling you to go do what I do. The right tool depends a lot on how you work, what other tools you already use, and what you need AI to actually do for you.
What I can do is lay all 4 side by side in plain English… and help you figure out which one fits you.
And here’s the first thing worth knowing: the price isn’t really the difference. The main paid plan for all 4 of these tools lands somewhere between $20 and $30 a month. If you’re picking based on price, it’s basically a coin flip.
The real differences are in what you get for that money. In the last post, I walked through the 8 building blocks every AI tool shares: custom instructions, memory, projects, skills, connectors, plugins, agents and scheduled tasks. That’s the lens I’m going to use here. Not benchmarks. Not who scored higher on some test. Just… what each tool gives you to actually get work done.
Let’s dive in.
The Quick Version
If you only read one section, make it this one.
| If you… | Look hardest at |
|---|---|
| Live in Gmail, Google Docs and Google Drive all day | Gemini |
| Mostly write, or want AI to do real work inside your business systems | Claude |
| Want the most features in one app, including image generation | ChatGPT |
| Live on X, want an easy “AI team,” and like a bundle | Grok |
| Want the cheapest paid option that’s still useful | Gemini ($4.99) or ChatGPT Go ($8) |
That’s not a ranking. Every one of these tools is good. They’re just good at different things, and they’re built by companies with different priorities.
Who Makes These, and What They’re Built Around
Quick background, because it explains a lot about how each tool feels.
- ChatGPT is from OpenAI. It’s the one most people tried first, and it’s built to do a little bit of everything for everybody.
- Claude is from Anthropic. It’s known for writing quality, and lately it’s been pushing hard into “agentic” work… AI that actually does the job instead of just talking about it.
- Gemini is from Google. It’s built to plug into the Google world: Gmail, Docs, Drive, Calendar, Photos.
- Grok is from Elon Musk’s AI company, tied closely to X (formerly Twitter). It has live access to what’s happening on X, which none of the others do.
What They Cost
Here’s the current lineup. US prices, per month.
A quick warning before you read the table. These prices and plan names are accurate as I write this (November 2026), but this stuff changes fast. Several of these companies changed their plans in just the last few weeks. Before you pull out your credit card, check the company’s own pricing page: ChatGPT, Claude, Gemini and Grok. Treat this table as a starting point, not gospel.
| Free | Budget | Main plan | Power user | Small team | |
|---|---|---|---|---|---|
| ChatGPT | Yes (with ads) | Go: $8 (with ads) | Plus: $20 | Pro: $100, $200 or $500 | Business: $25 per seat |
| Claude | Yes | None | Pro: $20 | Max: $100, Ultra: $200 | Team: $25 per seat |
| Gemini | Yes (lighter models) | AI Plus: $4.99 | AI Pro: $19.99 | AI Ultra: $99.99 or $199.99 | Google Workspace plans |
| Grok | Yes | Plus: $8 | Premium: $30 | Super: $100, Ultra: $200 | Teams, per seat |
A few things jump out.
ChatGPT now shows ads on its free and $8 plans. Nobody else puts ads in the assistant.
Gemini’s $19.99 plan includes 5 TB of Google storage. If you already pay Google for storage, that changes the math a lot.
Grok’s $30 plan is a bundle. It includes Grok Bot (more on that below), a Cursor Pro subscription (an AI coding tool), and X Premium+ perks. Grok also reorganized its plans in October. If you had an older plan (like my SuperGrok), you’re grandfathered in.
Claude’s Max and Ultra plans include API credits equal to what you pay. That matters mostly if you’re building things, but it’s a nice extra.
The usage limits nobody tells you about
Here’s something that trips people up. None of these companies tell you how many messages you get.
They all use some version of a “usage allowance.” It’s based on how much computing power your requests take, it resets every few hours, and there’s often a weekly cap on top. Ask a quick question and you barely touch it. Have it do a big research job or work through a long document, and you can burn through a lot.
They’re all a little vague about it. Claude gets the most complaints here, frankly. Its limits are the top gripe I see from users.
But here’s the thing. A lot of those complaints come down to one habit: using the biggest, most expensive model for everything.
Every one of these tools lets you pick which model you’re using, usually from a dropdown right in the chat box. The top models are amazing. They also eat through your allowance way faster than the everyday ones. Using Claude’s top model (Fable 5.2) to help you write a blog post is like taking a Lamborghini grocery shopping. Sure, it’ll get you there. It’s just overkill.
So learn what each model is good at, and pick the right one for the job. Save the heavy hitter for the hard stuff.
For me, Opus 5.5 is my go-to model in Claude for almost everything. It’s plenty smart for the kind of work I do, and it’s efficient enough that I’ve never hit my cap on the $100 Max plan… and I use Claude all day, all week long. I’d bet it’s the same story on the other tools. Run ChatGPT’s most advanced model on every little thing and you’ll burn through your plan. Dial it back to something that fits the task, and you’ll get a lot more out of the same money.
For most solopreneurs using AI a few times a day, the $20 to $30 plans are plenty on any of these tools… as long as you’re not driving the Lamborghini to the grocery store.
The 8 Building Blocks, Side by Side
This is the part that actually matters. Here’s how each tool handles the 8 building blocks from the last post.
| Building block | ChatGPT | Claude | Gemini | Grok |
|---|---|---|---|---|
| Custom instructions | ✅ | ✅ | ✅ (called “Instructions for Gemini”) | ✅ |
| Memory | ✅ | ✅ | ✅ | ✅ |
| Projects | ✅ | ✅ | ✅ (called “Notebooks”) | ✅ |
| Skills | ✅ | ✅ | ✅ (new, replacing “Gems”) | ✅ |
| Connectors | ✅ | ✅ | ✅ (deepest Google integration) | ✅ (smaller library) |
| Plugins | ✅ | ✅ | Developer tools only | Through Grok Bot |
| Agents | ✅ ChatGPT Work | ✅ Claude, Claude Code | ✅ Tasks in Gemini | ✅ Grok Bot |
| Scheduled tasks | ✅ | ✅ | ✅ | ✅ |
Notice something? It’s mostly checkmarks.
A year ago, this table would’ve looked very different. Now, all 4 tools have basically caught up to each other on the basics. They all remember you. They all connect to your other tools. They all have some kind of agent that can go do a job for you.
So the better question is how good each one is at a feature, and whether it fits the way you work.
Let me walk through the ones where the differences are real.
Memory: they all have it, but read the fine print
All 4 tools remember things between chats now. You can view and edit what they’ve remembered in all of them.
The catch is the same one I talked about in the last post. That memory lives inside their tool. If you switch, it doesn’t come with you. Claude does let you import and export memory, which is a nice touch. But as I said before, I wouldn’t build my business on any platform’s built-in memory. I keep mine in my own files, and point every tool at them.
Projects: same idea, different names
ChatGPT, Claude and Grok all have Projects… a workspace for each ongoing job with its own files and instructions.
Gemini has the same thing, it just calls them Notebooks. They sync with Gemini Notebook (which used to be called NotebookLM), and that’s genuinely good for working with a pile of your own documents. One catch: Notebooks aren’t available on Google Workspace accounts, which is exactly the kind of account a lot of businesses use.
Connectors: Gemini wins if you’re a Google person
Every tool connects to Gmail, Google Drive and a bunch of other services now. And most support MCP, which is the standard way to plug AI into just about anything.
But Gemini’s Google integration goes deeper, which makes sense. Google built it. It can work right inside Gmail and Docs (depending on your plan). If your business runs on Google Workspace, that’s a real advantage.
Grok’s connector library is smaller than the others. It covers the big ones (Gmail, Drive, Outlook, Notion, GitHub), and you can add your own, but there’s less to pick from.
Agents: this is where they really differ
All 4 have an agent now. They feel pretty different.
- ChatGPT Work builds documents, decks and websites, and browses the web for you. ChatGPT also has “always-on” agents on its Pro plans. And then there’s Codex, OpenAI’s answer to Claude Code (more on both below).
- Claude is merging its regular chat and its agent mode (Cowork) into one app. It started in September and is still rolling out. So you just ask, and it does the work. Then there’s Claude Code, which is what I use most. Despite the name, it’s not just for coding. It’s how I have AI work on my actual business: my WordPress sites, my files, my client records.
- Gemini’s agent is called Tasks in Gemini. It’s on the paid plans, and it isn’t available in the UK or Europe yet.
- Grok Bot is the most interesting one here, and it gets its own section below.
Each Tool, Plainly
Alright, so let’s go through each one. What it’s great at, where it falls short, and who it fits.
ChatGPT

For a lot of people, ChatGPT is AI. It’s the name everybody knows, the one most people tried first, and the one they think of by default. It’s kind of become the Google of AI.
And it’s earned that spot. It’s a very powerful platform. It’s just not the one I use.
ChatGPT is the Swiss Army knife. It does the most things in one place: chat, image generation, voice, web browsing, document and slide creation, even publishing simple web pages. It has add-ins for Microsoft Office and Google Sheets. Its free plan is generous.
Then there’s Codex. It’s basically OpenAI’s equivalent of Claude Code. It started out as a coding tool, but a lot of people now use it for serious agentic work… having AI actually work on their files and projects instead of just chatting about them. Free users get a small taste, and the paid plans get a lot more, with its own usage allowance. If agentic work is what you’re after, Codex is a big reason to take ChatGPT seriously.
The downsides?
It’s the most confusing of the bunch. There are a whole lot of model names, different allowances for different parts of the product, and the help pages sometimes contradict each other. The free and $8 plans have ads now.
And it changes constantly. This year alone, OpenAI retired its Atlas browser, its Canvas feature and its old agent mode. Custom GPTs (which a lot of people built their workflows around) are being retired on December 11. If you built something on a custom GPT, now’s the time to move it.
ChatGPT fits you if you want one app that does a little of everything, you want to make images, you want agentic work through Codex, or you’re already used to it and don’t feel like switching.
Claude

This is the one I use. So let me tell you why… and then tell you the downsides, because there are some.
Claude’s writing is the best of the bunch. That’s not just me saying it; it consistently comes out on top in blind writing tests. It sounds less like a robot right out of the gate.
But the bigger reason I use it is agentic work. Claude Code is how AI actually works on my business. It updates my WordPress sites, manages files, drafts content in my voice, keeps my records up to date. When I compare it side by side with Grok on that kind of work, it’s not close. Claude is way better at it.
Now, Claude actually gives you two ways to do agentic work, and it’s worth knowing the difference.
Cowork is the agent built into the regular Claude app. You ask, and it goes and does the job. It’s very powerful, and for a lot of people it’s all they’ll ever need. But it works inside the guardrails Anthropic sets for it. Every so often, it’ll get hung up on a security limitation and stop short of what you actually wanted it to do.
Claude Code works differently. It runs commands right on my own computer. So it can do pretty much anything I could do myself on that machine: work with my files, connect to my servers, run my tools. Where Cowork might hit a wall, Claude Code can usually just get it done. For the way I run my business, that’s the difference that matters. (That kind of power also means you want to know what you’re asking it to do. It’s working on your actual computer, after all.)
And here’s the part a lot of people get wrong… They hear “Claude Code” and assume it’s just for programmers. Or that you have to type commands into some scary black terminal window.
Not anymore. I use Claude Code entirely from the Claude desktop app (and, in my case, from the chat window in Hermes too). It looks and feels like a normal chat. You type what you want in plain English, just like you would in any AI tool. The difference is that it’s a chat that can actually reach out and do the work on your computer.
Anthropic also created a couple of the open standards the whole industry now uses (MCP for connectors, and the Agent Skills format). Both are now looked after by an independent foundation, not Anthropic. That matters to me because it means my setup isn’t locked into one company. A skill I build works in other tools too.
Now the downsides:
- No built-in image generation. Claude can’t make images on its own. But that doesn’t mean you need to sign up for a whole different AI tool. You can connect Claude to an image generation service, and it’ll happily make images for you. I have Claude Code connected to fal.ai and Replicate (two services that run image models), plus my Canva account. It generates images for me all day long. It just takes a little setup.
- Usage limits. This is the #1 complaint, and it’s fair. Heavy users hit the walls. But like I said earlier, a lot of that comes from running the biggest model on everything. Pick the right model for the job, and your plan goes a lot further.
- The best model costs extra. On the $20 plan, Claude’s top model (Fable) is only available through paid credits. Honestly, though? For most things, you don’t need Fable at all. Opus and Sonnet handle the vast majority of everyday work just fine.
- It changes a lot too. Plans and features get reshuffled every few months.
One more thing worth knowing. Anthropic announced in August that it now puts an invisible watermark in Claude’s writing. Most people won’t notice or care. But if you publish AI-written content without editing it, it’s worth knowing it can be identified. (I’ll come back to that in a later post about AI and disclosure.)
Claude fits you if writing quality matters to you, or you want AI to actually do work inside your business and not just answer questions.
Gemini

I’ll be honest… I don’t use Gemini much. And it’s not for lack of access.
I have a Google Workspace account, so I get the level of Gemini that comes with the standard Workspace plan. It works nicely inside Gmail and Docs, and Gemini has its own web interface too. I’ve just never found it nearly as useful as the other tools. It still feels pretty baked into the Google ecosystem… and I don’t live my entire life out of Google.
But if you do? If your business runs on Google, it’s hard to ignore.
It’s built into Gmail, Docs and Drive. The $19.99 plan includes 5 TB of storage. It handles a huge amount of text at once, so you can hand it long documents. It’s strong at images and video. And its $4.99 AI Plus plan is the cheapest real entry point of the 4.
The downsides:
- Reviews on its writing are mixed. It’s less consistent than the others.
- The free plan got weaker. In October, Google cut free users down to its lighter models.
- The names keep changing. Spark became Tasks in Gemini. Gems are becoming Skills (Gems stop working December 17). NotebookLM became Gemini Notebook. It’s a lot to keep up with.
- Some features want more of your data. A lot of Gemini’s best personal features depend on letting it into your Gmail and Photos, and on keeping your activity history turned on.
Gemini fits you if you live in Google Workspace and want AI right where you already work… or if you want the cheapest paid option.
Grok

I like Grok. I really do. I use it for personal stuff, brainstorming, research and searching. I talk to it by voice in the car a lot… its voice mode is really good.
What makes Grok different from the others is X. It has live access to what’s happening on X, which no other tool has. If you want to know what people are saying about something right now, Grok is the one to ask.
The $30 plan is also the best bundle of the 4. You get Grok, Grok Bot, a Cursor Pro subscription, and X Premium+ in one price. It also comes with an app builder (Grok Build) on every plan, even the free one.
Now, who uses Grok? I’ll share a gut hunch here. I’ve never had anybody tell me this, so take it for what it’s worth. But I suspect some people choose Grok partly because they trust Elon Musk more than they trust the other AI companies. And honestly? That’s as valid a reason as any. Every one of these tools comes from a company with its own priorities, and who you trust matters.
The downsides:
- It’s a great second tool, but a weaker main one. That’s both my experience and the general take in reviews. For serious work, the others pull ahead.
- A smaller connector library than the others.
- Every image and video it makes gets a watermark, and you can’t turn it off.
- Plan churn. It just switched to a new subscription system, so right now there are two sets of plans floating around.
- Safety controversies. Grok has had some ugly ones, including a wave of AI-generated sexual deepfakes that regulators in the UK and Europe are still investigating as I write this. I’m not going to pile on, but it’s fair for you to know about it.
A closer look at Grok Bot
Grok Bot deserves its own section, because it’s the most approachable version of something I think a lot of solopreneurs will want: an AI team.
Grok Bot gives you named AI “bots” that stay on all the time, with a cloud computer to work on. You can give them skills, set up routines that run on a schedule (up to 50 per bot), have several bots work together, and (on the team plans) put them in Slack or Microsoft Teams. It’s on the $30 plan and up, and it came out of beta in November.
That cloud computer is a big selling point. But Grok Bot can also run things directly on your own computer through its desktop app, similar to Claude Code or Cowork. And honestly, when I use Grok Bot, that’s the only way I use it. Depending only on the cloud computer gets a little annoying. I’d rather it work where my stuff actually is.
I use it for some agentic work. It’s basically like having a team. And it’s really user-friendly… it’s the easy version of what I do with a more technical setup.
But I don’t put my core business on it. Here’s why.
Grok Bot is a team you rent. Even when it’s working on my computer, the bots themselves, their setup, and everything they know live on someone else’s servers. If the plan changes, the price goes up, or the company decides to go a different direction, your “team” goes with it. For me, that’s a digital sovereignty problem. I’m fine renting the AI model itself. I’m not fine renting the whole operation.
There’s also a practical concern if you lean on the cloud side. All of your bots share the same cloud computer. They aren’t walled off from each other. So if you give one bot your login for a client’s site, and another bot handles something else, they’re all in the same space.
For my own setup, I use something called Hermes. It does a similar job, but it runs on infrastructure I own, using Claude as its brain. It’s nerdier… a lot nerdier. I’ll write about building your own AI team in a future post.
Grok fits you if you’re on X a lot, you want live information, you like the value of the bundle, or you want to try an AI team without getting technical.
A Quick Word on Privacy
This one surprises people.
ChatGPT, Gemini and Grok train on your chats by default on their personal plans. Claude asks you to choose when you sign up. Whichever one you use, you can turn it off, and it takes about 30 seconds:
- ChatGPT: Settings → Data controls → “Improve the model for everyone”
- Claude: Settings → Privacy → the model training setting
- Gemini: It’s tied to your “History” setting (it used to be called “Keep Activity”)
- Grok: Settings → Data Controls → “Improve the model”
If you work with client information, here’s a simple tip: the business and team plans don’t train on your data by default on any of these tools. A single seat on a team plan costs a few bucks more than the personal plan, and you get better privacy for it.
The Part Nobody Talks About: Everything Keeps Changing
Here’s what struck me most while researching this.
Every single one of these tools renamed, retired or reshuffled major features this year. ChatGPT is killing custom GPTs. Gemini is replacing Gems. Grok changed its whole subscription system. Claude merged two of its products into one and renamed its top plan. Some of what I’ve written in this post will probably be out of date in a few months.
So, how do you protect yourself from that?
You keep the important stuff outside the tool. Your business knowledge, your processes, your writing style… keep it in your own files. Use skills built on open standards where you can, so they move with you. Then the AI tool becomes something you can swap out, not something you’re stuck with.
That’s exactly why I keep my AI’s memory in my own files. When a tool changes, or a better one comes along, I just point the new one at the same files. It already knows me.
These Aren’t the Only Games in Town
These 4 are the best-known options here in the US, and they’re where I’d point most people to start. But they’re not the only ones.
There’s a whole world of “open” AI models, including some very capable ones out of China like DeepSeek and Qwen. You can reach a lot of them through services like OpenRouter, often for less than these subscriptions cost. (If privacy matters to you, it’s worth knowing where your data goes before using any of the Chinese companies’ own apps.)
There are also tools that let you build your own AI assistant or team and plug in whichever model you want. Hermes, which I mentioned earlier, is one. OpenClaw is another one people have been talking about. And if you want to go all the way, you can run AI on your own hardware so nothing leaves your office. I do a bit of that myself.
All of that is more technical, and I’ll get into it later in this series. For now, just know the big 4 are a starting point… not the whole menu.
So, Which One Should You Pay For?
I’m not going to give you one answer. It really does depend on you.
But before we get to how I’d choose, I want to push back on how most people think about this decision.
Stop looking at the price tag
Too many people get hung up on the monthly price. They look at $20, or $100, or $200 a month, and that’s the whole decision. How much money goes out the door.
What they’re not thinking about is how much work they can actually get done.
I’m on the Claude Max plan. That’s $100 a month. I know a lot of people would look at that and think it’s a super expensive plan.
Frankly? I look at it as chump change compared to what I get out of it.
I’ve got Claude so dialed in now that it’s like having an IT person working side by side with me. If I went out and hired that person, it would probably cost me $60,000 to $70,000 a year. I’m paying $1,200.
The amount of work you can get out of these tools is off the charts compared to what they cost. So look at it the way you’d look at any other business expense. What does it cost, and what does it return? Don’t try to cheap out. If your AI plan isn’t paying for itself several times over, something’s wrong with how you’re using it.
How I’d go about choosing
Start free. All 4 have free plans. Take one real task from your week (an email you need to write, a blog post outline, a pile of notes to sort through) and give it to each of them. You’ll feel the difference pretty quickly.
Then pay for one. Pick the one that fit best and get on a paid plan. The free plans hold back a lot. You won’t really know what a tool can do until you’re on a paid plan and setting up those building blocks. The main $20 to $30 plan is a fine place to start. As you hand it more real work and it starts earning its keep, move up.
Don’t pay for all of them. Seriously. Pick one main tool and get good at it. Add a second only when you hit something the first can’t do. For me, that’s how Grok ended up as my second tool… it does a few things Claude doesn’t, and I enjoy using it.
As for me? Claude is my main tool and Grok is my sidekick. That’s what works for how I run my business. Yours might look completely different, and that’s fine.
In the end, the tool matters less than how you set it up and what you have it do. Treat it like a chat box and even the $20 plan feels expensive. Put it to work on your business and even the $100 plan is a bargain.


