My Pro Is More Pro Than Your Pro

A dark interface showing an AI effort slider from Light and Pro to AGI, ending at "This Changes Everything."

The companies building superintelligence still haven’t figured out what to call it. And it has already driven me nuts.

Meta is retiring fruit codenames. They realized that Avocado, Mango, and Watermelon weren’t dignified enough for humanity’s next great leap.

Their current AI lineup includes Muse Wonder Full, because, you know, Wonder isn't full enough, and the marketing intern thought the wonder-ful wordplay is cute.

Google has Nano Banana. Anthropic has Haiku, Sonnet, Opus, Fable, and Mythos. So we've covered the supermarket and the library.

Of course, we only have OpenAI to blame for fucking up the industry's naming instincts. ChatGPT must be the worst successful product name ever. But OpenAI didn't stop with the product name. Like, it released GPT-4.1 after GPT-4.5, then recently expanded into Sol, Terra, Luna, and Astra. Perhaps Sam Altman didn't want to leave outer space to Elon Musk.

There's also Anthropic's Max and Ultracode, or whatever each company calls its top two effort levels. If you assume Max > Ultracode, because “maximum” kind of suggests we’ve reached the maximum, you'd be wrong. But notice how they added "code" to Ultra—they didn't want to confuse us with their own Ultra subscription tier.

And subscriptions are another can of worms. Anthropic used to offer two versions of Max. Now its subscriptions go Pro → Max → Ultra. Obviously, those are very different from Anthropic's own Max and Ultracode effort levels, and from OpenAI's Extra High and Ultra effort levels.

OpenAI, meanwhile, has three Pro plans. Even after you’ve gone Pro, you still have to decide how Pro you are, duh! Do you want 10×, 20× or 50× the allowance of Plus? Plus itself comes with an allowance of "15–160 messages per five hours." 15 to 160! How useful is a range like that? They could have just said "from something to something else."

For almost a month this year, OpenAI’s lineup was Astra 6, Sol 6.1, Terra 5.6, and Luna 6. And the newer, supposedly lower-tier Luna could outperform the older, higher-tier Terra on some tasks, at a fraction of the price. Same with Sol 6.1: it performs better than the bigger and costlier Astra 6 in some tasks. So remembering the hierarchy isn't enough; you need to factor in the naming convention, the release history, and the model card's benchmarks.

The only safe bet is that they're going to disrupt the hierarchy again. You think I am joking? Polymarket recorded $121,828 in trading volume on what Meta would name its next AI model family. The winning answer was "Other," and you could've doubled your money making that bet.

You might think this is an American thing, and while it started that way, the whole world got infected.

The French have Mistral, and Mistral has Ministral, Magistral, Pixtral, Devstral, Codestral, and Voxtral. I am so glad I'm not a student that gets quizzed on these. I've never used Mistral, so luckily I don't care to understand the difference between a Devstral and a Codestral.

And of course, we can't leave out China. Alibaba gave developers Qwen3-235B-A22B-Thinking-2507, which my password manager rates as a Very Strong password—one that would take 3 billion trillion years to crack. And DeepSeek released V3.1-Terminus, then V3.2-Exp seven days later, so even the end point came with a sequel.

It’s not like the AI labs aren’t aware. They very much are. They just don't care. In April 2025, Sam Altman acknowledged OpenAI deserved the mockery and said it would “fix our model naming by this summer.” Two summers later, it’s gotten worse.

A product name should help me understand what I’m buying, how it differs from the other options, and why I’d choose it.

Clever names are fine. Nano Banana is fun and memorable. But a naming system should teach me its logic once, then keep its promises. I don't want to do the branding department’s homework.

But given how much they keep switching strategies, I suspect we may have to wait for the first coming of superintelligence to put an end to this naming mess.

Subscribe to the Blog

me@example.com
Subscribe

Get in Touch