I think people have fatigue from the constant incremental upgrades to Chinese models (most of which are impressive on benchmarks). Also GLM isn’t a big name compared to Deepseek or even Kimi.
Yes, I think it’s fatigue. There have been so many incremental developments from Chinese models, and GLM 5.1 wasn’t an important model, so people ignored GLM 5.2 out of habit.
The 5.2 version number was a very bad choice by Z.ai.
I almost suspect that they deliberately chose a small version increase, to pretend to be anti-hype. It’s like the CEO who refuses to wear a suit, and instead dresses like a random guy on the street to prove he’s so high status he doesn’t even need a suit to show it. But when he meets with the investors, they don’t realize this and actually dismiss him as a random guy on the street. Whoops.
PS: I admit I’m still unsure how much benchmaxxing they did. It doesn’t look like benchmaxxing since they did better in software engineering than question answering, but you can never rule it out.
I think people have fatigue from the constant incremental upgrades to Chinese models (most of which are impressive on benchmarks). Also GLM isn’t a big name compared to Deepseek or even Kimi.
Yes, I think it’s fatigue. There have been so many incremental developments from Chinese models, and GLM 5.1 wasn’t an important model, so people ignored GLM 5.2 out of habit.
The 5.2 version number was a very bad choice by Z.ai.
I almost suspect that they deliberately chose a small version increase, to pretend to be anti-hype. It’s like the CEO who refuses to wear a suit, and instead dresses like a random guy on the street to prove he’s so high status he doesn’t even need a suit to show it. But when he meets with the investors, they don’t realize this and actually dismiss him as a random guy on the street. Whoops.
PS: I admit I’m still unsure how much benchmaxxing they did. It doesn’t look like benchmaxxing since they did better in software engineering than question answering, but you can never rule it out.