ok bear with me, since I have dyscalculia and don't get these things easily:
does the whole diagram say: "over time the model numbers change" and that's it?

Yes, it's the height is based on the vendor's model number they chose themselves at that time

So if I released some shitty Super Smarto 2000.2 tomorrow, I'd be in the lead

thank you for explaining, wasn't sure if I was being stupid or this diagram!

(but I'm gonna name my model "your name plus always one count higher than yours" so I'll win, ÄTSCH)

Joke's on them, I'm almost done training FaceBot 9000.

Granted, it's a 1.2B model and the training data is entirely My Little Pony fanfics. It's insane and thinks it's Princess Luna. But it's going to dominate this benchmark.

You can only tie with my AI, Ultra autocomplete: Version Infinity.

Gotta look past the bad charting - GPT's line is just going lower and lower with each release.

Number goes up, can't argue with that!

Oh sweet I always wanted to talk to Princess Luna!

Your avatar image checks out. :)

Nuh uh my AI is in version ℵ1

Its actually based on the Nightelves but yeah also applicable to best Princess :)

We don't want to lose the number war, but we didn't start it^TM^

Graph is wrong, GPT is clearly using negative numbers

My AI is version supercompact!

Even taking the chart at face value Anthropic looks on pace to catch up