Zai launches GLM-5.2 with a 1M-token context window and plans an MIT-licensed open-source release

GLM and Kimi are on the third iteration of their flagships. They too are in the fast update cycle.

GLM-5.2 supports two thinking-effort levels: High and Max.

For coding tasks, we recommend using Max effort to enable deeper reasoning and more reliable performance.

Obviously DeepSeek tends to ignore these dynamics, but I wish to see how much ground 4.1 covers in one step. Their objective, after all, was explicitly "fast iterations", and it's been 7 weeks since V4-Preview.

@Zai_org it should be open

yet only available via “GLM Coding Plan”? seems contradictory

no weights or even at least API

@Zai_org Oh God, what if this benchmark is actually real? We’re not ready.

@thegenioo @Zai_org r e a d

It's a good approach to releasing...

Launch Saturday to avoid disruption to businesses relying on the service weekdays; mid-day launch local time to have engineers monitoring; released to Coding Plan first for testing & feedback; encourage people to subscribe to run benchmarks.

@BennettBuhner @Zai_org wasn’t in the post 👀

@xeophon I mean they won't be matching anyone else's pace V4.1 comes out when they feel ready, not when Kimi K2.7 or GLM 5.2 drop. they may make some nod to the Fable situation but I doubt even that

@teortaxesTex > Obviously DeepSeek tends to ignore these dynamics

how many V3 iterations did we get? safe to say they were the first

@thegenioo @Zai_org open source next week

@BennettBuhner @Zai_org damn I think I was really sleepy lol wtf

@Zai_org Glad to hear you're launching for Coding Plan before API + router, this is the best way to test it. Releases getting smoother each time...

@teortaxesTex GLM 5.2 has a 1M context window too so i highkey wanna see if it hallucinates like DeepSeek does

@Zai_org a rush release that doesnt even have basic benchmark numbers - but a great timing due to the fable 5 drama. oss model go!!

@thegenioo @Zai_org Patience, they’re starting with coding plan first :)

@TotallyNotParth I wonder if that's possible to address in post-training alone

@thegenioo @Zai_org Hahaha 😂 it hit you harder than a frying pan I guess!

@Zai_org It’d be even better if it supported multimodal!

@BennettBuhner @Zai_org fable hit me hard like wtf man