OpenAI's Wild Ride: What the Heck is Going On with GPT-4o, GPT-4.5, and GPT-5?

It’s been a chaotic couple of weeks in AI land, and OpenAI has been at the center of it all.

Elon Musk tried (and failed) to buy OpenAI’s non-profit wing, then dropped Grok 3 like a mic, claiming it outperforms OpenAI’s own models. Claude Sonnet one-upped OpenAI’s software engineer benchmark. And yet—despite the turbulence—OpenAI is still absolutely crushing the AI game.

Let’s break it down.

The Numbers Don’t Lie: OpenAI is Winning

While the AI world loves drama, OpenAI is too busy raking in wins. According to COO Brad Lightcap:

• ChatGPT now has 400 million weekly active users.

• 2 million+ businesses use ChatGPT for work.

• Reasoning Model API usage shot up 5x since o3-mini launched.

• Oh, and when GPT-5 drops? It’ll be free.

Yep, you read that right. OpenAI is making GPT-5 totally free, with no usage caps.

So, what’s the catch?

Pro and Plus users will still have an advantage—they’ll get more inference time for complex tasks. OpenAI is also testing a slider scale (like Anthropic’s) to let users adjust power levels more intuitively.

But Wait… What’s Up With GPT-4o?

While everyone’s waiting for GPT-4.5 and GPT-5, OpenAI quietly slipped in an update to GPT-4o. And something’s… different.

• It’s smarter and refuses fewer requests.

• Many specialized custom GPTs (like AI illustrators) broke completely.

• Ethan Mollick’s popular Innovation GPT (10K+ users) stopped working entirely.

• Text-based GPTs? Still fine.

• OpenAI gave zero warning or documentation about this change.

What does this mean? If you’re a dev building on OpenAI’s tech, you’re in for a wild ride. Some businesses are already jumping ship to open-source alternatives like DeepSeek or Llama to avoid these unexpected model shake-ups.

GPT-4.5 and GPT-5: The Next Big Leap

According to The Verge, Microsoft is already prepping GPT-4.5 and GPT-5 to go live on Azure as we speak.

So, where does that leave us? OpenAI needs to figure out how to roll out massive upgrades without breaking the products of the 2M+ businesses relying on them. That’s the double-edged sword of working with OpenAI—cutting-edge innovation, but unpredictable stability.

One final twist: OpenAI quietly dropped a major Model Spec update on February 12th. Could this be the missing link to understanding the GPT-4o changes?

We’ll keep watching. In the meantime, buckle up—things are about to get interesting.