ChatGPT, Claude, and Gemini have switched directions six times

A new version of ChatGPT just dropped. A few days later, Claude. A few days after that, Gemini.

Honestly, doesn't it barely register anymore? Same here. I'd just shrug and think, “cool, it got smarter again.”

But after using these things every day, I started noticing something odd. They're not just getting smarter. They keep changing what they compete on. Six times so far.

1. Just be smart first (2023)

At first it was simple. Who's smarter? That was the whole game.

When GPT-4 came out in March 2023, everyone was kind of stunned. It passed the bar exam, wrote code, even got the jokes. Google (Bard back then) and Claude were scrambling to catch up.

The fight was about one thing: a higher score on a harder test.

2. It grew eyes and ears (early 2024)

But text alone hit a wall pretty fast. So the next thing was eyes and ears.

In May 2024, GPT-4o showed up and started seeing photos and hearing voices. Gemini was built from the ground up to handle text, images, and sound together.

Now you can snap a photo of your fridge and ask, “what can I make with this?” It started looking past the words.

3. It stopped forgetting (2024)

Third came memory.

Old AI was like someone who forgets what you said five minutes into a call. The longer the chat got, the more it lost track and went off the rails.

In February 2024, Gemini 1.5 arrived able to hold a whole book in its head at once. Claude could swallow a 200-page document in one go too.

So now you can drop in a long contract and say, “find the clauses that work against me.” No more freezing up in front of long documents.

4. Hold on, let me think (late 2024 to 2025)

The fourth one surprised me. It wasn't a smarter model. It was a slower one.

Old AI would open its mouth the second you asked, saying whatever came to mind. And wow, it said wrong answers with so much confidence.

Starting with OpenAI's o1 in September 2024, Claude and Gemini began saying “give me a sec” and working it out in their head first. Suddenly they got way more accurate at math and code, the stuff where one wrong step ruins everything.

The shift: from answering fast to answering right.

5. Not telling you how. Doing it for you.

This is the biggest shift. In one line: from “telling you” to “doing it for you.”

Before, if you asked “how do I find a flight?”, it told you how to search. Like a cookbook. The recipe is friendly, but you still have to cook.

Now you say “go find me a flight,” and it digs through the sites itself. Then it comes back with “how about this one?” The cookbook became your personal chef.

In October 2024, Claude went a step further. It started looking at the screen and clicking the mouse itself. First time I saw it, I got a little scared, honestly. We still don't know if this turns into Jarvis or Ultron.

6. Cheap and reliable is what actually wins (now)

The sixth one isn't flashy. But it might be the most important. Being cheap and not screwing up beats being smart.

Picture a genius who occasionally blows things up and costs a fortune. Now picture an average employee who just gets it done every time and is cheap to keep. When you're running a company, the second one wins.

That's why all three put out small, fast, cheap models: “mini,” “Flash,” “Haiku.” What actually ships in real products is mostly these.

So here's what I see

Line up all six shifts and a direction shows up. It started as “how smart is it,” and slowly moved to “can I trust it with the work.” From a test-score race to a good-employee race.

But here's the interesting part. That “can I trust it” part barely shows up in benchmark scores. There's a difference only the people building with these every day get to feel.

That's what the next post is about.