2 mystery models:

1) Zai's GLM multimodal flash model that already serves as much as DSV4-Flash on OpenCode 2) Kimi 3.1 - K3 from what I hear didn't have much post training Moonshot continuing its path of large models while Zai flexing its post training + distillation Ox Alpha is the bigger deal here. If you have a flash model @ V4-Flash cost that is beating all the publicly available closed models, what is the pt of Ant & OpenAI? We should all keep our eyes out on long Tang Jie posts, becaus — via @tphuang

Zai's GLM matches DSV4-Flash usage; Moonshot's K3 lacks post-training, signaling competitive AI shifts.