ChatGPT shamelessly copied Claude.

Sol-5.6 vs. Fable-5. The full breakdown.

ChatGPT finally copied Claude:

The new ChatGPT Work is just like Claude Cowork.

They announced:

• A new mega-app for ChatGPT.

• Codex, their coding tool, is now inside ChatGPT.

• A new mode inside the ChatGPT app, called ChatGPT Work.

• The new flagship model (5.6) with three levels of intelligence: Sol (very smart), Terra, and Luna (very cheap).

This entire newsletter will answer one question:

Does ChatGPT still matter?

---

Yes, it matters.

1. OpenAI is one of the two leading AI labs in the world.

2. ChatGPT will always matter, with its 5 billion monthly visits:

Top 10 websites by total visits in June 2026. ChatGPT is 5th. Claude is 37th.

But does it matter enough to switch from Claude?

Harder question. Let’s find out.

1. How to access the new ChatGPT.

You need to:

1. Uninstall the old ChatGPT app (if you had it before).

2. Download it (again, I know) at chatgpt.com/download.

3. You must have a paid account (at least $20/month).

It will look like this once you open it:

Top left: you can choose between Work and Codex.

If you click on “Chat” it opens a regular ChatGPT chat.

Now that you have access to the new ChatGPT, it’s time for you to know the truth.

2. It’s so confusing, I’m lost.

These are the 30 combinations of the new ChatGPT.

ChatGPT Work? Codex? Which model? Which effort? Good luck.

I saw this on Twitter (@rasbt), and I felt it. It’s so confusing.

Let me answer as simply as possible.

The first distinction is Work vs. Codex:

• ChatGPT Work is to work on anything but code. Like Claude Cowork, it spawns many ChatGPT (agents) to answer your queries.

• ChatGPT Codex is for developers. It’s the same, but with a design and an interface made for developers.

So you want to stay on ChatGPT Work.

Now, the second one is the different models:

• You have 5.5 (old one) and 5.6 (the new one).

• 5.6 comes in 3 flavors of intelligence…

  • Luna: cheap but dumb.

  • Terra: the cost/smartness balance.

  • Sol: the super smart super expensive one.

If that was it, that would be so easy… but no… you also have different efforts.

When you are on the selector (bottom right), it asks you which model, which effort, and which speed (faster = more expensive).

It’s quite obvious that the more effort, the smarter the model, but the more expensive it is. What’s not obvious at all is which one is better between

• 5.6 Sol (the smartest one) on Light effort?

• or 5.6 Luna (the dumbest one) on Extra High effort?

• and also, WHY DO WE HAVE TO PICK??? I can’t believe it.

So I tried to make a graph to explain it better… but it’s also a bit confusing. I’m sorry. I don’t work at OpenAI or Anthropic. I also wish it were easier.

Only use Luna-High, Terra-Ultra, or Sol-Ultra.

I’ll save you some thinking time:

• The best expensive one: Sol-Ultra, only for planning and a couple of turns.

• The one you should always use besides Sol-Ultra is Luna-High.

• I rarely used Terra-Ultra because it’s hard to know the “middle tasks”.

Hopefully, I cleared up a bit of the confusion around the new ChatGPT, the new models, and the new capabilities. It’s all fun and games, but IS IT WORTH IT???

3. “You’re out of Codex and Work usage”.

I pay for ChatGPT Business & Claude Business plans.

Why? Because it’s my job to test everything :) I also pay for Grok, Gemini & I regularly test DeepSeek, Kimi, Z.ai, Perplexity, NotebookLM.

Well, during my test of ChatGPT Work… I had almost no work done.

Because I kept seeing this:

You will see this screen very, very often.

“You’re out of Codex and Work usage.”

I honestly don’t understand why.

I can’t get anything done…

I am not the only one being very frustrated on X.

It went like this:

• I planned the task once with Sol-Ultra.

• Then I used Luna-High to do the tasks (just like I use Claude Fable-5).

• And I’m out of usage… in the middle of the task (!!!).

Even writing this newsletter was a challenge.

I couldn’t get anything done without hitting my limits.

Then I thought: “OK, maybe it’s because I pay $20 for ChatGPT, but I pay $100 for Claude. Let’s try to upgrade my subscription.”

You literally CANNOT get a better ChatGPT subscription.

It’s only “Business” with terribly low usage or “Enterprise” (but it’s minimum 150 seats).

On the flip side, Claude lets you pay $100/month if you want to:

I pay for Claude Premium seat. And I never hit my limits.

OK, but let’s say you want to switch to “ChatGPT Enterprise” to avoid hitting your limits. How bad is it? Let’s do a simulation with:

1. A company of 150 people (the minimum for an enterprise ChatGPT).

2. 10 new chats per workday per employee (so 10 x 5 days x 150 employees).

3. 10 turns per chat (you ask a question + ChatGPT answers = 2 turns).

4. Let’s say the first prompt is Sol (Ultra) and then only Luna (High).

5. The total cost of subscription is already $20 x 150 = $3,000 per month.

6. But the added cost of the new ChatGPT is… $17,000 per month.

And if you only use Sol (High), it’s $37,000 per month (or $250/employee).

I consult companies on their AI adoption.

I can’t imagine pitching “so it’s $20 per employee per month + $100-500 if they use it correctly and understand token management”. That would be crazy.

It’s much easier to say “it’s $100 per month per employee, virtually unlimited usage (besides Fable-5, which isn’t necessary for non-coding tasks).”

I offer discovery calls for enterprises of over 100 employees that need faster AI adoption (or any AI adoption at all). Send me a message for more info:

Now, to be extremely clear, Anthropic (Claude) also aggressively charges you beyond 150 seats. It goes like this:

• If you are at 150 Anthropic (Claude) seats, you can pay $20/$100 per seat and get virtually unlimited chats with Opus models. That’s what I do and recommend companies do.

• If you are at 151 Anthropic (Claude) seats, you pay per use (on top of a $20/seat subscription). The pricing becomes pretty much the same as ChatGPT. So $3,000/mo of subscription and easily tens of thousands of token costs on top of it…

It’s almost as if you should stay under the 150-seat mark. And I’m not the only one who can do basic maths:

This company went from $400k/year to $1.4M/year. So +1M/year, oof.

Token-maxxing is coming to an end. You can’t keep up once you pay for every token.

Token-maxxing is how companies want their employees to use as much AI tokens as possible (so agressively using AI).

A quick recap:

• Claude up to 150 seats is a no-brainer because of the $100/mo tier for businesses. You have a lot of free usage inside the $100 tier.

• Claude & ChatGPT are neck-and-neck past 150 seats, and super expensive, too.

So can you still use ChatGPT for work?

4. Can you still use ChatGPT for work?

I never really stopped my ChatGPT subscription.

First, because I can’t. It’s my freaking job aha. Over 800,000 people are reading my newsletters every week, so I better be on top of AI, always.

Second, because I love ChatGPT for its image generation. It is still the best on the market, by very far. Most of my newsletter covers are made first on Midjourney and then edited on ChatGPT. Maybe it’s time for me to short Adobe.

OK, now about the new models (especially Sol, the smartest one) and ChatGPT Work (the Cowork killer), this is what I won’t tell you about them:

1. They are smart and capable of doing similar things to Claude.

2. You can make PDFs, edit files on your computer, make Excel…

3. You can vibecode the same way Claude does.

4. The difference is very slim. Too slim to care.

You are not reading a newsletter to know what you can do with it; you want to know what’s different about the model & the new ChatGPT mode.

Let’s only talk about the differences:

#1: ChatGPT has templates.

When you click on “Plugins,” you can find templates:

Click here.

Or here.

So far, again, nothing too different.

I gave it a task, and it took a whopping 16 minutes to complete it on Standard (not the Fast mode, which is 1.5x more expensive). I edited the full video:

The result? I was happier with the Claude (Fable-5) version than the ChatGPT (Sol-5.6) version:

Claude is more direct, to the point. It’s not adding fluff. It took 13 minutes.

As always, the best test is YOU. Use a personal ChatGPT account, download the app, and check for your case study which one is best. I can’t answer it better than you. I mean, I can; this is what my consulting firm does for businesses.

You can also create your own templates inside ChatGPT. I guess that’s a bit different from Claude:

But then what’s the difference between a template and a skill? I don’t know; you tell me. Well, actually, I asked ChatGPT:

So template for text, skills for heavier stuff, I guess.

#2: ChatGPT has Skills.

Claude too. But let’s test the ones from ChatGPT:

You go to “Plugins” and then “Skills”.

Good news: you can transfer any Claude skills to ChatGPT.

Go to “Create a skill”.

Upload your Claude skill with this prompt:

Prompt: make a skill from this skill, just update "Claude" to "ChatGPT"

And I can now see it!

Nothing (very) different from Claude.

I tried to make it different, I promise.

#3: ChatGPT has Sites.

This time it’s really different!! It’s a new tab:

ChatGPT hyped it up. Let’s try it together.

I instantly wanted to try it on important matters, like solving cancer:

I use AI for good, I know.

I now have to choose from 3 different design directions. I picked the first.

It took something like 15 minutes from start to finish, but I’m happy with the result.

Problem: I can only share it internally because it’s my business ChatGPT account… a shame. But let’s open it.

I must sign in with ChatGPT… and you can also see how it burned my tokens. OK, let’s continue still.

It was cool. A fun experience. You can imagine many more capabilities in the future, kind of like Lovable (and the other vibecoding tools).

But for now, ChatGPT Sites is not as good as ChatGPT Codex or Claude Code to make your entire website from start to finish.

As AI gets better, it will be closer to ChatGPT Sites than to coding (you don’t see any “coding” element anywhere but only the end result, right inside the internet).

5. So ChatGPT or Claude?

Claude.

As you have already understood, they are very similar.

But my company has <150 employees, so the $100/month tier is way, waaaay better than the $20/mo tier from ChatGPT. We pay for a single shared ChatGPT account ($20/month) for image generation.

What if you have over 150 people in your company?

Well, it sucks. I would look into ways to stay under the 150 mark. 151 employees (and over) make AI very expensive to run with closed labs (Anthropic/OpenAI).

OK, but ChatGPT or Claude?

Over 150 employees, it only depends on what your team is comfortable with.

Your entire team loves Claude? They should stay there.

Your entire team only understands ChatGPT? Go there.

Because the real opportunity cost is not Claude vs. ChatGPT.

It’s paying for AI your team doesn’t touch. Adoption is the whole game.

When my team consults businesses, they never lack knowledge on how to use a tool. There is always a Superman of AI internally, the AI champion. The problem is how do you get 150+ Supermans, and can you even get there?

My core belief is that some people love the technology. They love it so much that they will constantly push themselves to try, test, and get better at it (here, AI). They don’t count time or energy on it. Kind of like someone collecting stamps. They just love the game.

Your entire team won’t be AI champions.

Just like your entire team is not the best at Excel. But they should be able to 1/ open it 2/ understand it 3/ work with it 4/ edit it if needed.

Who pays you, Claude or ChatGPT?

None.

I want to help millions of people know how to use AI. I can’t be paid by the AI labs I talk about every single day. One because it would be low-trust for you. Second, because I would need approval for every word.

I write constantly. I test everything. I can’t be stopped by contracts.

I do have partnerships with Gamma, Granola & Wispr Flow: I love their slides, meeting notes and dictation tools respectively. You don’t have to use them though.

But which AI am I using every single day? That’s a bigger one. So I use Claude. For now. And I am ready to switch when the jump is so big & obvious I must.

This newsletter is free, and will remain free. It grows from your shares, which are also free. I am forever grateful for your 2-second support that keeps How to AI free:

I do train businesses, teams of over 50 people, on whichever tool stack you guys are using. The goal is to get stronger & faster adoption without getting a million-dollar token bill.

Send me a message if you want a discovery call with my partners:

---

You have recommendation on my next newsletter?

Leave a comment, and my AI will compile all of them:

---

Upgrade your subscription to join our Circle community: