Zhipu releases open-weights GLM-5.2 with a 1-million-token context window for agentic coding

@jietang you guys need to do nothing, they are burning themselves.

@jietang Aren’t your model subject to censorship by the CCP?

How is that not the same as being a state regulated lab?

So China can have regulations you abide by but, the other countries cannot have their own laws.

Very honest and very fair.

Stop astroturfing openness and freedom.

@Geero_Gero @jietang Chinese models are censored, but they are open source and available for everyone to tinker with. Frontier American models are censored too, but they aren’t open sourced and now the government won’t even let you use them.

America is the more restrictive country here.

@jietang this is the way. will be offering GLM in OpenGateway

@jietang The point is, you can't ask for 8964 in GLM. But who fucking cares? Use it. 16x agents swarm, and code. By the way, if you still want to ask for 8964 as an open-source model, you can remove its safeguard entirely.

@jietang MiniMax and now you guys both saying "this is the most powerful open source model yet" while not actually releasing it open source for who knows how long it is blatently false advertising to say its open weights when it is in fact not and only will be """soon(tm)"""

I got to meet some of your team in Singapore at SuperAI. As someone that spent most of my time coding with western models, I gotta say that I am more excited than ever to use GLM. @ZixuanLi_ is awesome brilliant guy.

Ps. Thank you for the stickers! All the 42 references are appreciated!

@jietang Sure lets give every evil state actor, terrorist and hacker on the earth access to AI that will allow them to exploit systemic vulnerabilities world-wide.

@Geero_Gero @jietang Talking as if American models don't have propaganda injected into them lmao

@sambam_at @jietang China is pro open source because it helps them control that industry, it also helps divert earnings from US AI labs and route data to China that would otherwise train US models, it isn’t altruism and they will change the tone when the context changes.

@jietang Bro shit is getting Crazy this was one of those warning shots to pivot to local ai before we getting fully cucked

@jietang Thank you Jietang. Wise words, and could never be more true.

I hope that my fellow Americans can read this and understand the value of this, that China and Chinese people are our allies and friends.

We should be working together to advance humanity.

@jietang GLM 5.2 busy working in my Farnsworth Loop parallel agent best of n Claude Code dynamic workflow 🚀

@jietang But you guys are distilling from US closed source models. You could not do what you're doing without them.

@jietang long-horizon tasks matter most, keeping agent state stable over a million tokens beats raw context size every time

@jietang Please it is time for big model 🙏 we need something between 3T ~5T.

@jietang @loktar00 Incredible timing. I have been using GLM 5.1 a lot in Droid, it’s such a great model, can’t wait to try 5.2!

@jietang open release timed against a closed-model restriction is the loudest possible move. wonder how much of the demand pull comes from the past 24h, not the eval gains.

@jietang People are sleeping on the Chinese AI Labs