Two Drops, One Tuesday

Alibaba shipped Wan today. Open source. Free. A video generation model that, if you had described it to anyone in 2022, they would have assumed required a data center, a nine-figure investment, and a three-year waitlist — now you can just download it and run it, which is the kind of sentence that has stopped landing correctly because we've said it so many times.

The "completely wild" reflex is still there. You feel it. But the system has adapted. The shock gets processed and filed and you move on because there's another one.

ZhipuAI dropped GLM-4.5 today. Same day. Z.ai. Seemingly outta nowhere, which is the other reflex — the one where you look up from whatever you were doing and realize something substantial arrived without a six-week hype cycle or a countdown page or a partnership announcement with a major US cloud provider. It's just there. Here are the numbers. Here is the model. Goodbye.

This is the thing that I think is genuinely hard to sit with: the tempo. Not the quality of any individual release — the quality is obviously high, that part is obvious, that part has been obvious for a while — but the pace at which "significant" has been drained of meaning by sheer repetition.

Two major model releases, same Tuesday, one from a company that makes shopping carts and cloud infrastructure, one from a Chinese AI lab most people in the US couldn't have named six months ago. Both free. Both legitimately competitive with things that cost money.

The reaction in my head, and probably yours, was something like: huh. Not: this changes everything. Not: the world will never be the same. Just — huh. Let me look at the benchmarks for a second. Okay. That's pretty good. Anyway.

I don't think this means we've become numb to progress — or not only that. I think it means the word "wild" has done too many reps. It was already fatigued. We kept using it on things that were genuinely surprising and now the word itself is broken, lying on the floor, unable to distinguish between "a video model that would have won an Oscar three years ago" and "a really good YouTube thumbnail."

The benchmarks on GLM-4.5 are real. The Wan open source release is real. Both matter.

It's Tuesday.