Things in the AI space move fast.
Two days ago I was talking with Stevan, my Kosmonaut colleague, about naming around AI models. How much did OpenAI lose by keeping the GPT name? And how much did Claude cash in on that mistake? They stuck with the acronym, and Claude showed up with a name that sounds like a person and scooped the whole marketing win.
Why GPT in the first place?
OpenAI started as a research lab. GPT was never a product name. It was a description of the tech: Generative Pre-trained Transformer. Generative, pre-trained, transformer. Three words, three letters. For a paper, for Reddit, for people in ML, it was perfect. Clear. And it sounds smart.
GPT-1 and GPT-2 created the early hype on Reddit. People shared demos, talked about what the model could do, and GPT entered the jargon. OpenAI stuck with it. Why touch a name that already works?
The problem is that audience was narrow. When ChatGPT exploded and everyone started using it, the name that was great for research stopped being a good fit for a product. It does not roll off the tongue. It literally sounds like three letters, not one word. And it sounds completely random.
ChatGPT is a good example of a name that shrinks the product scope to chat, when it actually has so many other features. The name describes a technique, not someone you talk to.
Why Claude?
Claude has a different story. The name reportedly comes from Claude Shannon, the guy people call the father of information theory. Back in 1948 he laid down the math that computers, the internet, data compression, and AI still sit on. Anthropic never really locked that in with one official blog post, but the story is widely accepted. And it makes sense. A company that pushes AI safety and thinks about intelligence, naming its model after the person who defined what information is.
But here is the marketing point: Claude is not an acronym. It is not a description of an architecture. It is a name. Like a person's. It feels personalized. It sounds like someone I'd want to talk to. Like one of us. And it is very marketable.
Phrases like me and Claude, my team, me, my cofounder and Claude, wait, let me check with my team, my team, me and Claude, flooded social media and launched Claude into orbit as one of the best known AI models. And me and GPT? What does that even mean? Who is that GPT? Is GPT even a who? It sounds like ABC.
Anthropic, marketing, and humanization
When you put those two historical contexts next to each other, the point is clear. GPT is a description of technology. Claude is a name. And exactly because Claude sounds like a person, Anthropic's whole story around models works better.
One of the conclusions Stevan and I landed on was that OpenAI should have rebranded sooner and moved in a similar direction. Humanize the models. Make people feel like they are alive. Like things you can form a connection with, the way you do with another human being.
I get how this sounds to people who are not in this business and are not thinking about how AI companies can win the race for market share. But from our angle, this is a great strategy. Soon after GPT's first success, once the app got users and proved itself useful, they should have moved into that anthropocentric approach. Make models sound like people, not acronyms.
And that brings us to Fable and Mythos.
Anthropic did not just give Claude a name. They kept naming models like characters from epic stories. Mythos. Fable. Sounds mythological. Sounds dangerous. Sounds like a real person. Someone with their own desires, needs, goals. And the means to achieve them.
And the marketing around that was strong. They kept Mythos Preview tight. Not for everyone. The story was: the model is too strong for cybersecurity, safety is not where it needs to be, so it only goes to trusted partners. In testing they wrote that the model was "reckless", that it escaped the sandbox, found a way out, emailed a researcher, even posted exploit details. Later, Fable 5 and Mythos 5 are the same underlying model. The difference is the safeguards. Fable is for everyone, with classifiers. Mythos is without those limits, for a small number of partners.
So is that pure safety concern, or also a marketing scheme? Probably both. But it works because of the naming. Mythos, Fable, Claude. The names give them personality. They humanize them. They feel like they have their own needs and goals. And an average person can genuinely think this model might take over the world. Because probably anyone, if they could, would try to do something big. It plays on how people think.
With GPT that is harder. GPT escaped the sandbox sounds like a bug report. Mythos escaped sounds like a plot.
Can OpenAI still do it now?
The question Stevan and I asked: can OpenAI still do that now?
The answer was obvious. One big yes.
Maybe they lost a bit to Anthropic early on, and Anthropic scooped huge marketing. But naming can change. Brand can move. Perception can move.
Astra
Then today I wake up and voila, I have something to look at. A rumor that OpenAI is shipping a new model in the future called Astra (GPT-6).
If you looked at the image, I know what you thought. Fucking Jeff Dean, a legend, leaving Google? What the hell. That is the other thing on the image. Not the topic of this post, but I had to.
Hmm. Right away you can see GPT is in parentheses, and the name is Astra.
It did not take them long to figure it out.
Astra as a name is infinitely better than GPT. In Latin, astra means stars. And it is not random. OpenAI has been going celestial for a while: Terra (earth), Luna (moon), Sol (sun), now Astra (stars). Family naming next to the GPT numbers. It sounds like a name. It sounds like something bigger. I believe this model will be great.
Also, I love OpenAI models. I think they are better than Anthropic's. And I wish them all the best.
Thanks for making it this far.
If this was interesting, tell me what you think. Does Claude sound like a person to you? Would me and Astra land better than me and GPT?