What's in a Name? The Wild, Weird History of AI Model Names

If you've been following the AI space at all (or just trying to figure out what tool to recommend to your team), you've probably noticed that the names are... a lot. GPT-4o. Claude Opus 4.6. Gemini 2.5 Flash. Llama 4 Maverick. What does any of this mean? Who's naming these things? And why does it feel like you need a decoder ring just to have a conversation about which chatbot to try?

I spend most of my time helping social services organizations make sense of AI tools, so I've had to learn this naming landscape inside and out. And honestly? Once you dig into it, the stories behind these names are pretty interesting. Some are genuinely clever. Some are baffling. And a few are just funny.

Let's break it down.

Real Scientists, Real Stories

Several AI companies named their models after real people who shaped computing and science.

Claude (Anthropic) is named after Claude Shannon, a mathematician widely considered the father of information theory. Shannon's 1948 paper on communication theory basically laid the groundwork for everything from cell phones to, well, the very AI models that now carry his name. There's also been commentary that "Claude" was a deliberate choice of a male-sounding name to counter the trend of feminine AI assistants like Siri and Alexa. It's a small but interesting branding decision.

OpenAI's original GPT-3 models took this a step further. Instead of one name, they had four: Ada, Babbage, Curie, and Davinci, ranked from smallest to most powerful. Each honored a legendary figure. Ada Lovelace, often called the world's first computer programmer. Charles Babbage, known as the father of the computer. Marie Curie, the only person to win Nobel Prizes in two different scientific fields. And Leonardo da Vinci, history's most famous polymath. The naming choice was a nice touch: a hierarchy of luminaries to match a hierarchy of model capability.

DALL-E: Art Meets Robots

OpenAI's image generator DALL-E deserves its own section because the name is genuinely clever. It's a mashup of Salvador Dalí, the surrealist painter, and WALL-E, the Pixar robot. Art meets technology, surrealism meets AI-generated images. It's one of the few AI names that actually tells you what the product does while also being memorable.

Haiku, Sonnet, and Opus

Anthropic's model tier names are where things get literary. Their three sizes, from smallest to largest, are Haiku, Sonnet, and Opus. A haiku is tiny and precise (17 syllables). A sonnet is longer, more structured, with room for complexity. An opus is a major work, the big production.

The metaphor actually maps pretty cleanly to how the models work. Haiku is fast and cheap, built for quick tasks. Sonnet balances quality and speed for everyday work. Opus is the powerhouse for deep reasoning and complex problems. It's one of the more logical naming systems in the AI space, even if it does require you to remember your high school English class.

Google's Zoo

Google has bounced around on naming more than anyone. Their PaLM 2 model (which powered the original Bard chatbot) used animal names for its four sizes: Gecko, Otter, Bison, and Unicorn, from smallest to largest. The idea was intuitive enough: a gecko is tiny and fast, a unicorn is the rare and powerful one.

Then Google launched Gemini (named after the zodiac sign for twins, hinting at dual capability) and switched the tier names to Nano, Pro, Ultra, and Flash. The animals were quietly retired. Nobody talked about Otter ever again.

Grok: Deep Understanding (Allegedly)

Elon Musk's xAI named their model Grok, which comes from Robert Heinlein's 1961 novel "Stranger in a Strange Land." In the book, "grok" means to understand something so deeply and intuitively that you essentially become one with it. It's a term that's been popular in programmer culture for decades. Whether the AI lives up to that lofty meaning is debatable, but the name definitely appeals to the tech crowd.

GPT, LLaMA, and Friends

Some names are just technical acronyms that became household words. GPT stands for Generative Pre-trained Transformer, which is a description of the model's architecture. Nobody calls it that, though. Everyone just says "GPT" or "ChatGPT." The "Chat" part was added to signal that this version was built for conversation, not just text completion.

Meta's LLaMA originally stood for "Large Language Model Meta AI." By version 2, they quietly dropped the acronym angle and just called it "Llama" with normal capitalization. Easier to say, easier to type, and it comes with a built-in cute mental image.

Mistral's "-stral" Universe

Mistral AI, the French company, went all in on a naming suffix. Their coding model is Codestral. Their vision model is Pixtral. Math? Mathstral. Small model? Ministral. Reasoning? Magistral. You can almost predict the next one. It's the AI equivalent of naming all your kids with the same first letter.

Where It All Falls Apart

Beyond the creative names, the versioning is where things really fall apart. GPT-4.1 was released after GPT-5. A model called 4.1 came after 5. It's a separate product line.

OpenAI skipped "o2" in their reasoning model line (o1 jumped to o3) because O2 is a European telecom brand. Claude's version numbers have gone 3, 3.5, 3.7, 4, 4.5, 4.6. Google had a model codenamed "nano-banana" show up on benchmarks, and their CEO responded on social media with nothing but banana emojis.

It's a mess, and it's only getting messier.

The Best, the Worst, and the Funniest

If I had to hand out awards:

Names might seem like trivial marketing decisions, but they actually affect how people relate to these tools. When nonprofit staff hear "Claude," it sounds like a person they could talk to. When they hear "GPT-4o-mini," it sounds like something they need an engineering degree to use. Naming shapes whether people feel invited in or locked out.

For organizations considering AI adoption, the name of the tool matters less than what it can do for your team. But understanding the landscape, including the weird naming conventions, helps you feel more confident when evaluating options. And confidence is half the battle when you're introducing something new.

The real skill isn't memorizing model names. It's learning to direct these tools well, regardless of what they're called. I wrote more about that idea in The New Leverage: Why the People Who Direct AI Will Win, and it's the core of what I help organizations build toward.